net/core/neighbour.c

   1 /*
   2  *      Generic address resolution entity
   3  *
   4  *      Authors:
   5  *      Pedro Roque             <roque@di.fc.ul.pt>
   6  *      Alexey Kuznetsov        <kuznet@ms2.inr.ac.ru>
   7  *
   8  *      This program is free software; you can redistribute it and/or
   9  *      modify it under the terms of the GNU General Public License
  10  *      as published by the Free Software Foundation; either version
  11  *      2 of the License, or (at your option) any later version.
  12  *
  13  *      Fixes:
  14  *      Vitaly E. Lavrov        releasing NULL neighbor in neigh_add.
  15  *      Harald Welte            Add neighbour cache statistics like rtstat
  16  */
  17
  18 #include <linux/types.h>
  19 #include <linux/kernel.h>
  20 #include <linux/module.h>
  21 #include <linux/socket.h>
  22 #include <linux/netdevice.h>
  23 #include <linux/proc_fs.h>
  24 #ifdef CONFIG_SYSCTL
  25 #include <linux/sysctl.h>
  26 #endif
  27 #include <linux/times.h>
  28 #include <net/net_namespace.h>
  29 #include <net/neighbour.h>
  30 #include <net/dst.h>
  31 #include <net/sock.h>
  32 #include <net/netevent.h>
  33 #include <net/netlink.h>
  34 #include <linux/rtnetlink.h>
  35 #include <linux/random.h>
  36 #include <linux/string.h>
  37 #include <linux/log2.h>
  38
  39 #define NEIGH_DEBUG 1
  40
  41 #define NEIGH_PRINTK(x...) printk(x)
  42 #define NEIGH_NOPRINTK(x...) do { ; } while(0)
  43 #define NEIGH_PRINTK0 NEIGH_PRINTK
  44 #define NEIGH_PRINTK1 NEIGH_NOPRINTK
  45 #define NEIGH_PRINTK2 NEIGH_NOPRINTK
  46
  47 #if NEIGH_DEBUG >= 1
  48 #undef NEIGH_PRINTK1
  49 #define NEIGH_PRINTK1 NEIGH_PRINTK
  50 #endif
  51 #if NEIGH_DEBUG >= 2
  52 #undef NEIGH_PRINTK2
  53 #define NEIGH_PRINTK2 NEIGH_PRINTK
  54 #endif
  55
  56 #define PNEIGH_HASHMASK         0xF
  57
  58 static void neigh_timer_handler(unsigned long arg);
  59 static void __neigh_notify(struct neighbour *n, int type, int flags);
  60 static void neigh_update_notify(struct neighbour *neigh);
  61 static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev);
  62
  63 static struct neigh_table *neigh_tables;
  64 #ifdef CONFIG_PROC_FS
  65 static const struct file_operations neigh_stat_seq_fops;
  66 #endif
  67
  68 /*
  69    Neighbour hash table buckets are protected with rwlock tbl->lock.
  70
  71    - All the scans/updates to hash buckets MUST be made under this lock.
  72    - NOTHING clever should be made under this lock: no callbacks
  73      to protocol backends, no attempts to send something to network.
  74      It will result in deadlocks, if backend/driver wants to use neighbour
  75      cache.
  76    - If the entry requires some non-trivial actions, increase
  77      its reference count and release table lock.
  78
  79    Neighbour entries are protected:
  80    - with reference count.
  81    - with rwlock neigh->lock
  82
  83    Reference count prevents destruction.
  84
  85    neigh->lock mainly serializes ll address data and its validity state.
  86    However, the same lock is used to protect another entry fields:
  87     - timer
  88     - resolution queue
  89
  90    Again, nothing clever shall be made under neigh->lock,
  91    the most complicated procedure, which we allow is dev->hard_header.
  92    It is supposed, that dev->hard_header is simplistic and does
  93    not make callbacks to neighbour tables.
  94
  95    The last lock is neigh_tbl_lock. It is pure SMP lock, protecting
  96    list of neighbour tables. This list is used only in process context,
  97  */
  98
  99 static DEFINE_RWLOCK(neigh_tbl_lock);
 100
 101 static int neigh_blackhole(struct sk_buff *skb)
 102 {
 103         kfree_skb(skb);
 104         return -ENETDOWN;
 105 }
 106
 107 static void neigh_cleanup_and_release(struct neighbour *neigh)
 108 {
 109         if (neigh->parms->neigh_cleanup)
 110                 neigh->parms->neigh_cleanup(neigh);
 111
 112         __neigh_notify(neigh, RTM_DELNEIGH, 0);
 113         neigh_release(neigh);
 114 }
 115
 116 /*
 117  * It is random distribution in the interval (1/2)*base...(3/2)*base.
 118  * It corresponds to default IPv6 settings and is not overridable,
 119  * because it is really reasonable choice.
 120  */
 121
 122 unsigned long neigh_rand_reach_time(unsigned long base)
 123 {
 124         return (base ? (net_random() % base) + (base >> 1) : 0);
 125 }
 126 EXPORT_SYMBOL(neigh_rand_reach_time);
 127
 128
 129 static int neigh_forced_gc(struct neigh_table *tbl)
 130 {
 131         int shrunk = 0;
 132         int i;
 133
 134         NEIGH_CACHE_STAT_INC(tbl, forced_gc_runs);
 135
 136         write_lock_bh(&tbl->lock);
 137         for (i = 0; i <= tbl->hash_mask; i++) {
 138                 struct neighbour *n, **np;
 139
 140                 np = &tbl->hash_buckets[i];
 141                 while ((n = *np) != NULL) {
 142                         /* Neighbour record may be discarded if:
 143                          * - nobody refers to it.
 144                          * - it is not permanent
 145                          */
 146                         write_lock(&n->lock);
 147                         if (atomic_read(&n->refcnt) == 1 &&
 148                             !(n->nud_state & NUD_PERMANENT)) {
 149                                 *np     = n->next;
 150                                 n->dead = 1;
 151                                 shrunk  = 1;
 152                                 write_unlock(&n->lock);
 153                                 neigh_cleanup_and_release(n);
 154                                 continue;
 155                         }
 156                         write_unlock(&n->lock);
 157                         np = &n->next;
 158                 }
 159         }
 160
 161         tbl->last_flush = jiffies;
 162
 163         write_unlock_bh(&tbl->lock);
 164
 165         return shrunk;
 166 }
 167
 168 static void neigh_add_timer(struct neighbour *n, unsigned long when)
 169 {
 170         neigh_hold(n);
 171         if (unlikely(mod_timer(&n->timer, when))) {
 172                 printk("NEIGH: BUG, double timer add, state is %x\n",
 173                        n->nud_state);
 174                 dump_stack();
 175         }
 176 }
 177
 178 static int neigh_del_timer(struct neighbour *n)
 179 {
 180         if ((n->nud_state & NUD_IN_TIMER) &&
 181             del_timer(&n->timer)) {
 182                 neigh_release(n);
 183                 return 1;
 184         }
 185         return 0;
 186 }
 187
 188 static void pneigh_queue_purge(struct sk_buff_head *list)
 189 {
 190         struct sk_buff *skb;
 191
 192         while ((skb = skb_dequeue(list)) != NULL) {
 193                 dev_put(skb->dev);
 194                 kfree_skb(skb);
 195         }
 196 }
 197
 198 static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev)
 199 {
 200         int i;
 201
 202         for (i = 0; i <= tbl->hash_mask; i++) {
 203                 struct neighbour *n, **np = &tbl->hash_buckets[i];
 204
 205                 while ((n = *np) != NULL) {
 206                         if (dev && n->dev != dev) {
 207                                 np = &n->next;
 208                                 continue;
 209                         }
 210                         *np = n->next;
 211                         write_lock(&n->lock);
 212                         neigh_del_timer(n);
 213                         n->dead = 1;
 214
 215                         if (atomic_read(&n->refcnt) != 1) {
 216                                 /* The most unpleasant situation.
 217                                    We must destroy neighbour entry,
 218                                    but someone still uses it.
 219
 220                                    The destroy will be delayed until
 221                                    the last user releases us, but
 222                                    we must kill timers etc. and move
 223                                    it to safe state.
 224                                  */
 225                                 skb_queue_purge(&n->arp_queue);
 226                                 n->output = neigh_blackhole;
 227                                 if (n->nud_state & NUD_VALID)
 228                                         n->nud_state = NUD_NOARP;
 229                                 else
 230                                         n->nud_state = NUD_NONE;
 231                                 NEIGH_PRINTK2("neigh %p is stray.\n", n);
 232                         }
 233                         write_unlock(&n->lock);
 234                         neigh_cleanup_and_release(n);
 235                 }
 236         }
 237 }
 238
 239 void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev)
 240 {
 241         write_lock_bh(&tbl->lock);
 242         neigh_flush_dev(tbl, dev);
 243         write_unlock_bh(&tbl->lock);
 244 }
 245 EXPORT_SYMBOL(neigh_changeaddr);
 246
 247 int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
 248 {
 249         write_lock_bh(&tbl->lock);
 250         neigh_flush_dev(tbl, dev);
 251         pneigh_ifdown(tbl, dev);
 252         write_unlock_bh(&tbl->lock);
 253
 254         del_timer_sync(&tbl->proxy_timer);
 255         pneigh_queue_purge(&tbl->proxy_queue);
 256         return 0;
 257 }
 258 EXPORT_SYMBOL(neigh_ifdown);
 259
 260 static struct neighbour *neigh_alloc(struct neigh_table *tbl)
 261 {
 262         struct neighbour *n = NULL;
 263         unsigned long now = jiffies;
 264         int entries;
 265
 266         entries = atomic_inc_return(&tbl->entries) - 1;
 267         if (entries >= tbl->gc_thresh3 ||
 268             (entries >= tbl->gc_thresh2 &&
 269              time_after(now, tbl->last_flush + 5 * HZ))) {
 270                 if (!neigh_forced_gc(tbl) &&
 271                     entries >= tbl->gc_thresh3)
 272                         goto out_entries;
 273         }
 274
 275         n = kmem_cache_zalloc(tbl->kmem_cachep, GFP_ATOMIC);
 276         if (!n)
 277                 goto out_entries;
 278
 279         skb_queue_head_init(&n->arp_queue);
 280         rwlock_init(&n->lock);
 281         n->updated        = n->used = now;
 282         n->nud_state      = NUD_NONE;
 283         n->output         = neigh_blackhole;
 284         n->parms          = neigh_parms_clone(&tbl->parms);
 285         setup_timer(&n->timer, neigh_timer_handler, (unsigned long)n);
 286
 287         NEIGH_CACHE_STAT_INC(tbl, allocs);
 288         n->tbl            = tbl;
 289         atomic_set(&n->refcnt, 1);
 290         n->dead           = 1;
 291 out:
 292         return n;
 293
 294 out_entries:
 295         atomic_dec(&tbl->entries);
 296         goto out;
 297 }
 298
 299 static struct neighbour **neigh_hash_alloc(unsigned int entries)
 300 {
 301         unsigned long size = entries * sizeof(struct neighbour *);
 302         struct neighbour **ret;
 303
 304         if (size <= PAGE_SIZE) {
 305                 ret = kzalloc(size, GFP_ATOMIC);
 306         } else {
 307                 ret = (struct neighbour **)
 308                       __get_free_pages(GFP_ATOMIC|__GFP_ZERO, get_order(size));
 309         }
 310         return ret;
 311 }
 312
 313 static void neigh_hash_free(struct neighbour **hash, unsigned int entries)
 314 {
 315         unsigned long size = entries * sizeof(struct neighbour *);
 316
 317         if (size <= PAGE_SIZE)
 318                 kfree(hash);
 319         else
 320                 free_pages((unsigned long)hash, get_order(size));
 321 }
 322
 323 static void neigh_hash_grow(struct neigh_table *tbl, unsigned long new_entries)
 324 {
 325         struct neighbour **new_hash, **old_hash;
 326         unsigned int i, new_hash_mask, old_entries;
 327
 328         NEIGH_CACHE_STAT_INC(tbl, hash_grows);
 329
 330         BUG_ON(!is_power_of_2(new_entries));
 331         new_hash = neigh_hash_alloc(new_entries);
 332         if (!new_hash)
 333                 return;
 334
 335         old_entries = tbl->hash_mask + 1;
 336         new_hash_mask = new_entries - 1;
 337         old_hash = tbl->hash_buckets;
 338
 339         get_random_bytes(&tbl->hash_rnd, sizeof(tbl->hash_rnd));
 340         for (i = 0; i < old_entries; i++) {
 341                 struct neighbour *n, *next;
 342
 343                 for (n = old_hash[i]; n; n = next) {
 344                         unsigned int hash_val = tbl->hash(n->primary_key, n->dev);
 345
 346                         hash_val &= new_hash_mask;
 347                         next = n->next;
 348
 349                         n->next = new_hash[hash_val];
 350                         new_hash[hash_val] = n;
 351                 }
 352         }
 353         tbl->hash_buckets = new_hash;
 354         tbl->hash_mask = new_hash_mask;
 355
 356         neigh_hash_free(old_hash, old_entries);
 357 }
 358
 359 struct neighbour *neigh_lookup(struct neigh_table *tbl, const void *pkey,
 360                                struct net_device *dev)
 361 {
 362         struct neighbour *n;
 363         int key_len = tbl->key_len;
 364         u32 hash_val;
 365
 366         NEIGH_CACHE_STAT_INC(tbl, lookups);
 367
 368         read_lock_bh(&tbl->lock);
 369         hash_val = tbl->hash(pkey, dev);
 370         for (n = tbl->hash_buckets[hash_val & tbl->hash_mask]; n; n = n->next) {
 371                 if (dev == n->dev && !memcmp(n->primary_key, pkey, key_len)) {
 372                         neigh_hold(n);
 373                         NEIGH_CACHE_STAT_INC(tbl, hits);
 374                         break;
 375                 }
 376         }
 377         read_unlock_bh(&tbl->lock);
 378         return n;
 379 }
 380 EXPORT_SYMBOL(neigh_lookup);
 381
 382 struct neighbour *neigh_lookup_nodev(struct neigh_table *tbl, struct net *net,
 383                                      const void *pkey)
 384 {
 385         struct neighbour *n;
 386         int key_len = tbl->key_len;
 387         u32 hash_val;
 388
 389         NEIGH_CACHE_STAT_INC(tbl, lookups);
 390
 391         read_lock_bh(&tbl->lock);
 392         hash_val = tbl->hash(pkey, NULL);
 393         for (n = tbl->hash_buckets[hash_val & tbl->hash_mask]; n; n = n->next) {
 394                 if (!memcmp(n->primary_key, pkey, key_len) &&
 395                     net_eq(dev_net(n->dev), net)) {
 396                         neigh_hold(n);
 397                         NEIGH_CACHE_STAT_INC(tbl, hits);
 398                         break;
 399                 }
 400         }
 401         read_unlock_bh(&tbl->lock);
 402         return n;
 403 }
 404 EXPORT_SYMBOL(neigh_lookup_nodev);
 405
 406 struct neighbour *neigh_create(struct neigh_table *tbl, const void *pkey,
 407                                struct net_device *dev)
 408 {
 409         u32 hash_val;
 410         int key_len = tbl->key_len;
 411         int error;
 412         struct neighbour *n1, *rc, *n = neigh_alloc(tbl);
 413
 414         if (!n) {
 415                 rc = ERR_PTR(-ENOBUFS);
 416                 goto out;
 417         }
 418
 419         memcpy(n->primary_key, pkey, key_len);
 420         n->dev = dev;
 421         dev_hold(dev);
 422
 423         /* Protocol specific setup. */
 424         if (tbl->constructor && (error = tbl->constructor(n)) < 0) {
 425                 rc = ERR_PTR(error);
 426                 goto out_neigh_release;
 427         }
 428
 429         /* Device specific setup. */
 430         if (n->parms->neigh_setup &&
 431             (error = n->parms->neigh_setup(n)) < 0) {
 432                 rc = ERR_PTR(error);
 433                 goto out_neigh_release;
 434         }
 435
 436         n->confirmed = jiffies - (n->parms->base_reachable_time << 1);
 437
 438         write_lock_bh(&tbl->lock);
 439
 440         if (atomic_read(&tbl->entries) > (tbl->hash_mask + 1))
 441                 neigh_hash_grow(tbl, (tbl->hash_mask + 1) << 1);
 442
 443         hash_val = tbl->hash(pkey, dev) & tbl->hash_mask;
 444
 445         if (n->parms->dead) {
 446                 rc = ERR_PTR(-EINVAL);
 447                 goto out_tbl_unlock;
 448         }
 449
 450         for (n1 = tbl->hash_buckets[hash_val]; n1; n1 = n1->next) {
 451                 if (dev == n1->dev && !memcmp(n1->primary_key, pkey, key_len)) {
 452                         neigh_hold(n1);
 453                         rc = n1;
 454                         goto out_tbl_unlock;
 455                 }
 456         }
 457
 458         n->next = tbl->hash_buckets[hash_val];
 459         tbl->hash_buckets[hash_val] = n;
 460         n->dead = 0;
 461         neigh_hold(n);
 462         write_unlock_bh(&tbl->lock);
 463         NEIGH_PRINTK2("neigh %p is created.\n", n);
 464         rc = n;
 465 out:
 466         return rc;
 467 out_tbl_unlock:
 468         write_unlock_bh(&tbl->lock);
 469 out_neigh_release:
 470         neigh_release(n);
 471         goto out;
 472 }
 473 EXPORT_SYMBOL(neigh_create);
 474
 475 static u32 pneigh_hash(const void *pkey, int key_len)
 476 {
 477         u32 hash_val = *(u32 *)(pkey + key_len - 4);
 478         hash_val ^= (hash_val >> 16);
 479         hash_val ^= hash_val >> 8;
 480         hash_val ^= hash_val >> 4;
 481         hash_val &= PNEIGH_HASHMASK;
 482         return hash_val;
 483 }
 484
 485 static struct pneigh_entry *__pneigh_lookup_1(struct pneigh_entry *n,
 486                                               struct net *net,
 487                                               const void *pkey,
 488                                               int key_len,
 489                                               struct net_device *dev)
 490 {
 491         while (n) {
 492                 if (!memcmp(n->key, pkey, key_len) &&
 493                     net_eq(pneigh_net(n), net) &&
 494                     (n->dev == dev || !n->dev))
 495                         return n;
 496                 n = n->next;
 497         }
 498         return NULL;
 499 }
 500
 501 struct pneigh_entry *__pneigh_lookup(struct neigh_table *tbl,
 502                 struct net *net, const void *pkey, struct net_device *dev)
 503 {
 504         int key_len = tbl->key_len;
 505         u32 hash_val = pneigh_hash(pkey, key_len);
 506
 507         return __pneigh_lookup_1(tbl->phash_buckets[hash_val],
 508                                  net, pkey, key_len, dev);
 509 }
 510 EXPORT_SYMBOL_GPL(__pneigh_lookup);
 511
 512 struct pneigh_entry * pneigh_lookup(struct neigh_table *tbl,
 513                                     struct net *net, const void *pkey,
 514                                     struct net_device *dev, int creat)
 515 {
 516         struct pneigh_entry *n;
 517         int key_len = tbl->key_len;
 518         u32 hash_val = pneigh_hash(pkey, key_len);
 519
 520         read_lock_bh(&tbl->lock);
 521         n = __pneigh_lookup_1(tbl->phash_buckets[hash_val],
 522                               net, pkey, key_len, dev);
 523         read_unlock_bh(&tbl->lock);
 524
 525         if (n || !creat)
 526                 goto out;
 527
 528         ASSERT_RTNL();
 529
 530         n = kmalloc(sizeof(*n) + key_len, GFP_KERNEL);
 531         if (!n)
 532                 goto out;
 533
 534 #ifdef CONFIG_NET_NS
 535         n->net = hold_net(net);
 536 #endif
 537         memcpy(n->key, pkey, key_len);
 538         n->dev = dev;
 539         if (dev)
 540                 dev_hold(dev);
 541
 542         if (tbl->pconstructor && tbl->pconstructor(n)) {
 543                 if (dev)
 544                         dev_put(dev);
 545                 release_net(net);
 546                 kfree(n);
 547                 n = NULL;
 548                 goto out;
 549         }
 550
 551         write_lock_bh(&tbl->lock);
 552         n->next = tbl->phash_buckets[hash_val];
 553         tbl->phash_buckets[hash_val] = n;
 554         write_unlock_bh(&tbl->lock);
 555 out:
 556         return n;
 557 }
 558 EXPORT_SYMBOL(pneigh_lookup);
 559
 560
 561 int pneigh_delete(struct neigh_table *tbl, struct net *net, const void *pkey,
 562                   struct net_device *dev)
 563 {
 564         struct pneigh_entry *n, **np;
 565         int key_len = tbl->key_len;
 566         u32 hash_val = pneigh_hash(pkey, key_len);
 567
 568         write_lock_bh(&tbl->lock);
 569         for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL;
 570              np = &n->next) {
 571                 if (!memcmp(n->key, pkey, key_len) && n->dev == dev &&
 572                     net_eq(pneigh_net(n), net)) {
 573                         *np = n->next;
 574                         write_unlock_bh(&tbl->lock);
 575                         if (tbl->pdestructor)
 576                                 tbl->pdestructor(n);
 577                         if (n->dev)
 578                                 dev_put(n->dev);
 579                         release_net(pneigh_net(n));
 580                         kfree(n);
 581                         return 0;
 582                 }
 583         }
 584         write_unlock_bh(&tbl->lock);
 585         return -ENOENT;
 586 }
 587
 588 static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
 589 {
 590         struct pneigh_entry *n, **np;
 591         u32 h;
 592
 593         for (h = 0; h <= PNEIGH_HASHMASK; h++) {
 594                 np = &tbl->phash_buckets[h];
 595                 while ((n = *np) != NULL) {
 596                         if (!dev || n->dev == dev) {
 597                                 *np = n->next;
 598                                 if (tbl->pdestructor)
 599                                         tbl->pdestructor(n);
 600                                 if (n->dev)
 601                                         dev_put(n->dev);
 602                                 release_net(pneigh_net(n));
 603                                 kfree(n);
 604                                 continue;
 605                         }
 606                         np = &n->next;
 607                 }
 608         }
 609         return -ENOENT;
 610 }
 611
 612 static void neigh_parms_destroy(struct neigh_parms *parms);
 613
 614 static inline void neigh_parms_put(struct neigh_parms *parms)
 615 {
 616         if (atomic_dec_and_test(&parms->refcnt))
 617                 neigh_parms_destroy(parms);
 618 }
 619
 620 /*
 621  *      neighbour must already be out of the table;
 622  *
 623  */
 624 void neigh_destroy(struct neighbour *neigh)
 625 {
 626         struct hh_cache *hh;
 627
 628         NEIGH_CACHE_STAT_INC(neigh->tbl, destroys);
 629
 630         if (!neigh->dead) {
 631                 printk(KERN_WARNING
 632                        "Destroying alive neighbour %p\n", neigh);
 633                 dump_stack();
 634                 return;
 635         }
 636
 637         if (neigh_del_timer(neigh))
 638                 printk(KERN_WARNING "Impossible event.\n");
 639
 640         while ((hh = neigh->hh) != NULL) {
 641                 neigh->hh = hh->hh_next;
 642                 hh->hh_next = NULL;
 643
 644                 write_seqlock_bh(&hh->hh_lock);
 645                 hh->hh_output = neigh_blackhole;
 646                 write_sequnlock_bh(&hh->hh_lock);
 647                 if (atomic_dec_and_test(&hh->hh_refcnt))
 648                         kfree(hh);
 649         }
 650
 651         skb_queue_purge(&neigh->arp_queue);
 652
 653         dev_put(neigh->dev);
 654         neigh_parms_put(neigh->parms);
 655
 656         NEIGH_PRINTK2("neigh %p is destroyed.\n", neigh);
 657
 658         atomic_dec(&neigh->tbl->entries);
 659         kmem_cache_free(neigh->tbl->kmem_cachep, neigh);
 660 }
 661 EXPORT_SYMBOL(neigh_destroy);
 662
 663 /* Neighbour state is suspicious;
 664    disable fast path.
 665
 666    Called with write_locked neigh.
 667  */
 668 static void neigh_suspect(struct neighbour *neigh)
 669 {
 670         struct hh_cache *hh;
 671
 672         NEIGH_PRINTK2("neigh %p is suspected.\n", neigh);
 673
 674         neigh->output = neigh->ops->output;
 675
 676         for (hh = neigh->hh; hh; hh = hh->hh_next)
 677                 hh->hh_output = neigh->ops->output;
 678 }
 679
 680 /* Neighbour state is OK;
 681    enable fast path.
 682
 683    Called with write_locked neigh.
 684  */
 685 static void neigh_connect(struct neighbour *neigh)
 686 {
 687         struct hh_cache *hh;
 688
 689         NEIGH_PRINTK2("neigh %p is connected.\n", neigh);
 690
 691         neigh->output = neigh->ops->connected_output;
 692
 693         for (hh = neigh->hh; hh; hh = hh->hh_next)
 694                 hh->hh_output = neigh->ops->hh_output;
 695 }
 696
 697 static void neigh_periodic_timer(unsigned long arg)
 698 {
 699         struct neigh_table *tbl = (struct neigh_table *)arg;
 700         struct neighbour *n, **np;
 701         unsigned long expire, now = jiffies;
 702
 703         NEIGH_CACHE_STAT_INC(tbl, periodic_gc_runs);
 704
 705         write_lock(&tbl->lock);
 706
 707         /*
 708          *      periodically recompute ReachableTime from random function
 709          */
 710
 711         if (time_after(now, tbl->last_rand + 300 * HZ)) {
 712                 struct neigh_parms *p;
 713                 tbl->last_rand = now;
 714                 for (p = &tbl->parms; p; p = p->next)
 715                         p->reachable_time =
 716                                 neigh_rand_reach_time(p->base_reachable_time);
 717         }
 718
 719         np = &tbl->hash_buckets[tbl->hash_chain_gc];
 720         tbl->hash_chain_gc = ((tbl->hash_chain_gc + 1) & tbl->hash_mask);
 721
 722         while ((n = *np) != NULL) {
 723                 unsigned int state;
 724
 725                 write_lock(&n->lock);
 726
 727                 state = n->nud_state;
 728                 if (state & (NUD_PERMANENT | NUD_IN_TIMER)) {
 729                         write_unlock(&n->lock);
 730                         goto next_elt;
 731                 }
 732
 733                 if (time_before(n->used, n->confirmed))
 734                         n->used = n->confirmed;
 735
 736                 if (atomic_read(&n->refcnt) == 1 &&
 737                     (state == NUD_FAILED ||
 738                      time_after(now, n->used + n->parms->gc_staletime))) {
 739                         *np = n->next;
 740                         n->dead = 1;
 741                         write_unlock(&n->lock);
 742                         neigh_cleanup_and_release(n);
 743                         continue;
 744                 }
 745                 write_unlock(&n->lock);
 746
 747 next_elt:
 748                 np = &n->next;
 749         }
 750
 751         /* Cycle through all hash buckets every base_reachable_time/2 ticks.
 752          * ARP entry timeouts range from 1/2 base_reachable_time to 3/2
 753          * base_reachable_time.
 754          */
 755         expire = tbl->parms.base_reachable_time >> 1;
 756         expire /= (tbl->hash_mask + 1);
 757         if (!expire)
 758                 expire = 1;
 759
 760         if (expire>HZ)
 761                 mod_timer(&tbl->gc_timer, round_jiffies(now + expire));
 762         else
 763                 mod_timer(&tbl->gc_timer, now + expire);
 764
 765         write_unlock(&tbl->lock);
 766 }
 767
 768 static __inline__ int neigh_max_probes(struct neighbour *n)
 769 {
 770         struct neigh_parms *p = n->parms;
 771         return (n->nud_state & NUD_PROBE ?
 772                 p->ucast_probes :
 773                 p->ucast_probes + p->app_probes + p->mcast_probes);
 774 }
 775
 776 /* Called when a timer expires for a neighbour entry. */
 777
 778 static void neigh_timer_handler(unsigned long arg)
 779 {
 780         unsigned long now, next;
 781         struct neighbour *neigh = (struct neighbour *)arg;
 782         unsigned state;
 783         int notify = 0;
 784
 785         write_lock(&neigh->lock);
 786
 787         state = neigh->nud_state;
 788         now = jiffies;
 789         next = now + HZ;
 790
 791         if (!(state & NUD_IN_TIMER)) {
 792 #ifndef CONFIG_SMP
 793                 printk(KERN_WARNING "neigh: timer & !nud_in_timer\n");
 794 #endif
 795                 goto out;
 796         }
 797
 798         if (state & NUD_REACHABLE) {
 799                 if (time_before_eq(now,
 800                                    neigh->confirmed + neigh->parms->reachable_time)) {
 801                         NEIGH_PRINTK2("neigh %p is still alive.\n", neigh);
 802                         next = neigh->confirmed + neigh->parms->reachable_time;
 803                 } else if (time_before_eq(now,
 804                                           neigh->used + neigh->parms->delay_probe_time)) {
 805                         NEIGH_PRINTK2("neigh %p is delayed.\n", neigh);
 806                         neigh->nud_state = NUD_DELAY;
 807                         neigh->updated = jiffies;
 808                         neigh_suspect(neigh);
 809                         next = now + neigh->parms->delay_probe_time;
 810                 } else {
 811                         NEIGH_PRINTK2("neigh %p is suspected.\n", neigh);
 812                         neigh->nud_state = NUD_STALE;
 813                         neigh->updated = jiffies;
 814                         neigh_suspect(neigh);
 815                         notify = 1;
 816                 }
 817         } else if (state & NUD_DELAY) {
 818                 if (time_before_eq(now,
 819                                    neigh->confirmed + neigh->parms->delay_probe_time)) {
 820                         NEIGH_PRINTK2("neigh %p is now reachable.\n", neigh);
 821                         neigh->nud_state = NUD_REACHABLE;
 822                         neigh->updated = jiffies;
 823                         neigh_connect(neigh);
 824                         notify = 1;
 825                         next = neigh->confirmed + neigh->parms->reachable_time;
 826                 } else {
 827                         NEIGH_PRINTK2("neigh %p is probed.\n", neigh);
 828                         neigh->nud_state = NUD_PROBE;
 829                         neigh->updated = jiffies;
 830                         atomic_set(&neigh->probes, 0);
 831                         next = now + neigh->parms->retrans_time;
 832                 }
 833         } else {
 834                 /* NUD_PROBE|NUD_INCOMPLETE */
 835                 next = now + neigh->parms->retrans_time;
 836         }
 837
 838         if ((neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) &&
 839             atomic_read(&neigh->probes) >= neigh_max_probes(neigh)) {
 840                 struct sk_buff *skb;
 841
 842                 neigh->nud_state = NUD_FAILED;
 843                 neigh->updated = jiffies;
 844                 notify = 1;
 845                 NEIGH_CACHE_STAT_INC(neigh->tbl, res_failed);
 846                 NEIGH_PRINTK2("neigh %p is failed.\n", neigh);
 847
 848                 /* It is very thin place. report_unreachable is very complicated
 849                    routine. Particularly, it can hit the same neighbour entry!
 850
 851                    So that, we try to be accurate and avoid dead loop. --ANK
 852                  */
 853                 while (neigh->nud_state == NUD_FAILED &&
 854                        (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
 855                         write_unlock(&neigh->lock);
 856                         neigh->ops->error_report(neigh, skb);
 857                         write_lock(&neigh->lock);
 858                 }
 859                 skb_queue_purge(&neigh->arp_queue);
 860         }
 861
 862         if (neigh->nud_state & NUD_IN_TIMER) {
 863                 if (time_before(next, jiffies + HZ/2))
 864                         next = jiffies + HZ/2;
 865                 if (!mod_timer(&neigh->timer, next))
 866                         neigh_hold(neigh);
 867         }
 868         if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
 869                 struct sk_buff *skb = skb_peek(&neigh->arp_queue);
 870                 /* keep skb alive even if arp_queue overflows */
 871                 if (skb)
 872                         skb = skb_copy(skb, GFP_ATOMIC);
 873                 write_unlock(&neigh->lock);
 874                 neigh->ops->solicit(neigh, skb);
 875                 atomic_inc(&neigh->probes);
 876                 if (skb)
 877                         kfree_skb(skb);
 878         } else {
 879 out:
 880                 write_unlock(&neigh->lock);
 881         }
 882
 883         if (notify)
 884                 neigh_update_notify(neigh);
 885
 886         neigh_release(neigh);
 887 }
 888
 889 int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb)
 890 {
 891         int rc;
 892         unsigned long now;
 893
 894         write_lock_bh(&neigh->lock);
 895
 896         rc = 0;
 897         if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
 898                 goto out_unlock_bh;
 899
 900         now = jiffies;
 901
 902         if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
 903                 if (neigh->parms->mcast_probes + neigh->parms->app_probes) {
 904                         atomic_set(&neigh->probes, neigh->parms->ucast_probes);
 905                         neigh->nud_state     = NUD_INCOMPLETE;
 906                         neigh->updated = jiffies;
 907                         neigh_add_timer(neigh, now + 1);
 908                 } else {
 909                         neigh->nud_state = NUD_FAILED;
 910                         neigh->updated = jiffies;
 911                         write_unlock_bh(&neigh->lock);
 912
 913                         if (skb)
 914                                 kfree_skb(skb);
 915                         return 1;
 916                 }
 917         } else if (neigh->nud_state & NUD_STALE) {
 918                 NEIGH_PRINTK2("neigh %p is delayed.\n", neigh);
 919                 neigh->nud_state = NUD_DELAY;
 920                 neigh->updated = jiffies;
 921                 neigh_add_timer(neigh,
 922                                 jiffies + neigh->parms->delay_probe_time);
 923         }
 924
 925         if (neigh->nud_state == NUD_INCOMPLETE) {
 926                 if (skb) {
 927                         if (skb_queue_len(&neigh->arp_queue) >=
 928                             neigh->parms->queue_len) {
 929                                 struct sk_buff *buff;
 930                                 buff = __skb_dequeue(&neigh->arp_queue);
 931                                 kfree_skb(buff);
 932                                 NEIGH_CACHE_STAT_INC(neigh->tbl, unres_discards);
 933                         }
 934                         __skb_queue_tail(&neigh->arp_queue, skb);
 935                 }
 936                 rc = 1;
 937         }
 938 out_unlock_bh:
 939         write_unlock_bh(&neigh->lock);
 940         return rc;
 941 }
 942 EXPORT_SYMBOL(__neigh_event_send);
 943
 944 static void neigh_update_hhs(struct neighbour *neigh)
 945 {
 946         struct hh_cache *hh;
 947         void (*update)(struct hh_cache*, const struct net_device*, const unsigned char *)
 948                 = neigh->dev->header_ops->cache_update;
 949
 950         if (update) {
 951                 for (hh = neigh->hh; hh; hh = hh->hh_next) {
 952                         write_seqlock_bh(&hh->hh_lock);
 953                         update(hh, neigh->dev, neigh->ha);
 954                         write_sequnlock_bh(&hh->hh_lock);
 955                 }
 956         }
 957 }
 958
 959
 960
 961 /* Generic update routine.
 962    -- lladdr is new lladdr or NULL, if it is not supplied.
 963    -- new    is new state.
 964    -- flags
 965         NEIGH_UPDATE_F_OVERRIDE allows to override existing lladdr,
 966                                 if it is different.
 967         NEIGH_UPDATE_F_WEAK_OVERRIDE will suspect existing "connected"
 968                                 lladdr instead of overriding it
 969                                 if it is different.
 970                                 It also allows to retain current state
 971                                 if lladdr is unchanged.
 972         NEIGH_UPDATE_F_ADMIN    means that the change is administrative.
 973
 974         NEIGH_UPDATE_F_OVERRIDE_ISROUTER allows to override existing
 975                                 NTF_ROUTER flag.
 976         NEIGH_UPDATE_F_ISROUTER indicates if the neighbour is known as
 977                                 a router.
 978
 979    Caller MUST hold reference count on the entry.
 980  */
 981
 982 int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
 983                  u32 flags)
 984 {
 985         u8 old;
 986         int err;
 987         int notify = 0;
 988         struct net_device *dev;
 989         int update_isrouter = 0;
 990
 991         write_lock_bh(&neigh->lock);
 992
 993         dev    = neigh->dev;
 994         old    = neigh->nud_state;
 995         err    = -EPERM;
 996
 997         if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
 998             (old & (NUD_NOARP | NUD_PERMANENT)))
 999                 goto out;
1000
1001         if (!(new & NUD_VALID)) {
1002                 neigh_del_timer(neigh);
1003                 if (old & NUD_CONNECTED)
1004                         neigh_suspect(neigh);
1005                 neigh->nud_state = new;
1006                 err = 0;
1007                 notify = old & NUD_VALID;
1008                 goto out;
1009         }
1010
1011         /* Compare new lladdr with cached one */
1012         if (!dev->addr_len) {
1013                 /* First case: device needs no address. */
1014                 lladdr = neigh->ha;
1015         } else if (lladdr) {
1016                 /* The second case: if something is already cached
1017                    and a new address is proposed:
1018                    - compare new & old
1019                    - if they are different, check override flag
1020                  */
1021                 if ((old & NUD_VALID) &&
1022                     !memcmp(lladdr, neigh->ha, dev->addr_len))
1023                         lladdr = neigh->ha;
1024         } else {
1025                 /* No address is supplied; if we know something,
1026                    use it, otherwise discard the request.
1027                  */
1028                 err = -EINVAL;
1029                 if (!(old & NUD_VALID))
1030                         goto out;
1031                 lladdr = neigh->ha;
1032         }
1033
1034         if (new & NUD_CONNECTED)
1035                 neigh->confirmed = jiffies;
1036         neigh->updated = jiffies;
1037
1038         /* If entry was valid and address is not changed,
1039            do not change entry state, if new one is STALE.
1040          */
1041         err = 0;
1042         update_isrouter = flags & NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1043         if (old & NUD_VALID) {
1044                 if (lladdr != neigh->ha && !(flags & NEIGH_UPDATE_F_OVERRIDE)) {
1045                         update_isrouter = 0;
1046                         if ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) &&
1047                             (old & NUD_CONNECTED)) {
1048                                 lladdr = neigh->ha;
1049                                 new = NUD_STALE;
1050                         } else
1051                                 goto out;
1052                 } else {
1053                         if (lladdr == neigh->ha && new == NUD_STALE &&
1054                             ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) ||
1055                              (old & NUD_CONNECTED))
1056                             )
1057                                 new = old;
1058                 }
1059         }
1060
1061         if (new != old) {
1062                 neigh_del_timer(neigh);
1063                 if (new & NUD_IN_TIMER)
1064                         neigh_add_timer(neigh, (jiffies +
1065                                                 ((new & NUD_REACHABLE) ?
1066                                                  neigh->parms->reachable_time :
1067                                                  0)));
1068                 neigh->nud_state = new;
1069         }
1070
1071         if (lladdr != neigh->ha) {
1072                 memcpy(&neigh->ha, lladdr, dev->addr_len);
1073                 neigh_update_hhs(neigh);
1074                 if (!(new & NUD_CONNECTED))
1075                         neigh->confirmed = jiffies -
1076                                       (neigh->parms->base_reachable_time << 1);
1077                 notify = 1;
1078         }
1079         if (new == old)
1080                 goto out;
1081         if (new & NUD_CONNECTED)
1082                 neigh_connect(neigh);
1083         else
1084                 neigh_suspect(neigh);
1085         if (!(old & NUD_VALID)) {
1086                 struct sk_buff *skb;
1087
1088                 /* Again: avoid dead loop if something went wrong */
1089
1090                 while (neigh->nud_state & NUD_VALID &&
1091                        (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1092                         struct neighbour *n1 = neigh;
1093                         write_unlock_bh(&neigh->lock);
1094                         /* On shaper/eql skb->dst->neighbour != neigh :( */
1095                         if (skb->dst && skb->dst->neighbour)
1096                                 n1 = skb->dst->neighbour;
1097                         n1->output(skb);
1098                         write_lock_bh(&neigh->lock);
1099                 }
1100                 skb_queue_purge(&neigh->arp_queue);
1101         }
1102 out:
1103         if (update_isrouter) {
1104                 neigh->flags = (flags & NEIGH_UPDATE_F_ISROUTER) ?
1105                         (neigh->flags | NTF_ROUTER) :
1106                         (neigh->flags & ~NTF_ROUTER);
1107         }
1108         write_unlock_bh(&neigh->lock);
1109
1110         if (notify)
1111                 neigh_update_notify(neigh);
1112
1113         return err;
1114 }
1115 EXPORT_SYMBOL(neigh_update);
1116
1117 struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1118                                  u8 *lladdr, void *saddr,
1119                                  struct net_device *dev)
1120 {
1121         struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1122                                                  lladdr || !dev->addr_len);
1123         if (neigh)
1124                 neigh_update(neigh, lladdr, NUD_STALE,
1125                              NEIGH_UPDATE_F_OVERRIDE);
1126         return neigh;
1127 }
1128 EXPORT_SYMBOL(neigh_event_ns);
1129
1130 static void neigh_hh_init(struct neighbour *n, struct dst_entry *dst,
1131                           __be16 protocol)
1132 {
1133         struct hh_cache *hh;
1134         struct net_device *dev = dst->dev;
1135
1136         for (hh = n->hh; hh; hh = hh->hh_next)
1137                 if (hh->hh_type == protocol)
1138                         break;
1139
1140         if (!hh && (hh = kzalloc(sizeof(*hh), GFP_ATOMIC)) != NULL) {
1141                 seqlock_init(&hh->hh_lock);
1142                 hh->hh_type = protocol;
1143                 atomic_set(&hh->hh_refcnt, 0);
1144                 hh->hh_next = NULL;
1145
1146                 if (dev->header_ops->cache(n, hh)) {
1147                         kfree(hh);
1148                         hh = NULL;
1149                 } else {
1150                         atomic_inc(&hh->hh_refcnt);
1151                         hh->hh_next = n->hh;
1152                         n->hh       = hh;
1153                         if (n->nud_state & NUD_CONNECTED)
1154                                 hh->hh_output = n->ops->hh_output;
1155                         else
1156                                 hh->hh_output = n->ops->output;
1157                 }
1158         }
1159         if (hh) {
1160                 atomic_inc(&hh->hh_refcnt);
1161                 dst->hh = hh;
1162         }
1163 }
1164
1165 /* This function can be used in contexts, where only old dev_queue_xmit
1166    worked, f.e. if you want to override normal output path (eql, shaper),
1167    but resolution is not made yet.
1168  */
1169
1170 int neigh_compat_output(struct sk_buff *skb)
1171 {
1172         struct net_device *dev = skb->dev;
1173
1174         __skb_pull(skb, skb_network_offset(skb));
1175
1176         if (dev_hard_header(skb, dev, ntohs(skb->protocol), NULL, NULL,
1177                             skb->len) < 0 &&
1178             dev->header_ops->rebuild(skb))
1179                 return 0;
1180
1181         return dev_queue_xmit(skb);
1182 }
1183 EXPORT_SYMBOL(neigh_compat_output);
1184
1185 /* Slow and careful. */
1186
1187 int neigh_resolve_output(struct sk_buff *skb)
1188 {
1189         struct dst_entry *dst = skb->dst;
1190         struct neighbour *neigh;
1191         int rc = 0;
1192
1193         if (!dst || !(neigh = dst->neighbour))
1194                 goto discard;
1195
1196         __skb_pull(skb, skb_network_offset(skb));
1197
1198         if (!neigh_event_send(neigh, skb)) {
1199                 int err;
1200                 struct net_device *dev = neigh->dev;
1201                 if (dev->header_ops->cache && !dst->hh) {
1202                         write_lock_bh(&neigh->lock);
1203                         if (!dst->hh)
1204                                 neigh_hh_init(neigh, dst, dst->ops->protocol);
1205                         err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1206                                               neigh->ha, NULL, skb->len);
1207                         write_unlock_bh(&neigh->lock);
1208                 } else {
1209                         read_lock_bh(&neigh->lock);
1210                         err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1211                                               neigh->ha, NULL, skb->len);
1212                         read_unlock_bh(&neigh->lock);
1213                 }
1214                 if (err >= 0)
1215                         rc = neigh->ops->queue_xmit(skb);
1216                 else
1217                         goto out_kfree_skb;
1218         }
1219 out:
1220         return rc;
1221 discard:
1222         NEIGH_PRINTK1("neigh_resolve_output: dst=%p neigh=%p\n",
1223                       dst, dst ? dst->neighbour : NULL);
1224 out_kfree_skb:
1225         rc = -EINVAL;
1226         kfree_skb(skb);
1227         goto out;
1228 }
1229 EXPORT_SYMBOL(neigh_resolve_output);
1230
1231 /* As fast as possible without hh cache */
1232
1233 int neigh_connected_output(struct sk_buff *skb)
1234 {
1235         int err;
1236         struct dst_entry *dst = skb->dst;
1237         struct neighbour *neigh = dst->neighbour;
1238         struct net_device *dev = neigh->dev;
1239
1240         __skb_pull(skb, skb_network_offset(skb));
1241
1242         read_lock_bh(&neigh->lock);
1243         err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1244                               neigh->ha, NULL, skb->len);
1245         read_unlock_bh(&neigh->lock);
1246         if (err >= 0)
1247                 err = neigh->ops->queue_xmit(skb);
1248         else {
1249                 err = -EINVAL;
1250                 kfree_skb(skb);
1251         }
1252         return err;
1253 }
1254 EXPORT_SYMBOL(neigh_connected_output);
1255
1256 static void neigh_proxy_process(unsigned long arg)
1257 {
1258         struct neigh_table *tbl = (struct neigh_table *)arg;
1259         long sched_next = 0;
1260         unsigned long now = jiffies;
1261         struct sk_buff *skb, *n;
1262
1263         spin_lock(&tbl->proxy_queue.lock);
1264
1265         skb_queue_walk_safe(&tbl->proxy_queue, skb, n) {
1266                 long tdif = NEIGH_CB(skb)->sched_next - now;
1267
1268                 if (tdif <= 0) {
1269                         struct net_device *dev = skb->dev;
1270                         __skb_unlink(skb, &tbl->proxy_queue);
1271                         if (tbl->proxy_redo && netif_running(dev))
1272                                 tbl->proxy_redo(skb);
1273                         else
1274                                 kfree_skb(skb);
1275
1276                         dev_put(dev);
1277                 } else if (!sched_next || tdif < sched_next)
1278                         sched_next = tdif;
1279         }
1280         del_timer(&tbl->proxy_timer);
1281         if (sched_next)
1282                 mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1283         spin_unlock(&tbl->proxy_queue.lock);
1284 }
1285
1286 void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1287                     struct sk_buff *skb)
1288 {
1289         unsigned long now = jiffies;
1290         unsigned long sched_next = now + (net_random() % p->proxy_delay);
1291
1292         if (tbl->proxy_queue.qlen > p->proxy_qlen) {
1293                 kfree_skb(skb);
1294                 return;
1295         }
1296
1297         NEIGH_CB(skb)->sched_next = sched_next;
1298         NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1299
1300         spin_lock(&tbl->proxy_queue.lock);
1301         if (del_timer(&tbl->proxy_timer)) {
1302                 if (time_before(tbl->proxy_timer.expires, sched_next))
1303                         sched_next = tbl->proxy_timer.expires;
1304         }
1305         dst_release(skb->dst);
1306         skb->dst = NULL;
1307         dev_hold(skb->dev);
1308         __skb_queue_tail(&tbl->proxy_queue, skb);
1309         mod_timer(&tbl->proxy_timer, sched_next);
1310         spin_unlock(&tbl->proxy_queue.lock);
1311 }
1312 EXPORT_SYMBOL(pneigh_enqueue);
1313
1314 static inline struct neigh_parms *lookup_neigh_params(struct neigh_table *tbl,
1315                                                       struct net *net, int ifindex)
1316 {
1317         struct neigh_parms *p;
1318
1319         for (p = &tbl->parms; p; p = p->next) {
1320                 if ((p->dev && p->dev->ifindex == ifindex && net_eq(neigh_parms_net(p), net)) ||
1321                     (!p->dev && !ifindex))
1322                         return p;
1323         }
1324
1325         return NULL;
1326 }
1327
1328 struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1329                                       struct neigh_table *tbl)
1330 {
1331         struct neigh_parms *p, *ref;
1332         struct net *net;
1333
1334         net = dev_net(dev);
1335         ref = lookup_neigh_params(tbl, net, 0);
1336         if (!ref)
1337                 return NULL;
1338
1339         p = kmemdup(ref, sizeof(*p), GFP_KERNEL);
1340         if (p) {
1341                 p->tbl            = tbl;
1342                 atomic_set(&p->refcnt, 1);
1343                 p->reachable_time =
1344                                 neigh_rand_reach_time(p->base_reachable_time);
1345
1346                 if (dev->neigh_setup && dev->neigh_setup(dev, p)) {
1347                         kfree(p);
1348                         return NULL;
1349                 }
1350
1351                 dev_hold(dev);
1352                 p->dev = dev;
1353 #ifdef CONFIG_NET_NS
1354                 p->net = hold_net(net);
1355 #endif
1356                 p->sysctl_table = NULL;
1357                 write_lock_bh(&tbl->lock);
1358                 p->next         = tbl->parms.next;
1359                 tbl->parms.next = p;
1360                 write_unlock_bh(&tbl->lock);
1361         }
1362         return p;
1363 }
1364 EXPORT_SYMBOL(neigh_parms_alloc);
1365
1366 static void neigh_rcu_free_parms(struct rcu_head *head)
1367 {
1368         struct neigh_parms *parms =
1369                 container_of(head, struct neigh_parms, rcu_head);
1370
1371         neigh_parms_put(parms);
1372 }
1373
1374 void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1375 {
1376         struct neigh_parms **p;
1377
1378         if (!parms || parms == &tbl->parms)
1379                 return;
1380         write_lock_bh(&tbl->lock);
1381         for (p = &tbl->parms.next; *p; p = &(*p)->next) {
1382                 if (*p == parms) {
1383                         *p = parms->next;
1384                         parms->dead = 1;
1385                         write_unlock_bh(&tbl->lock);
1386                         if (parms->dev)
1387                                 dev_put(parms->dev);
1388                         call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
1389                         return;
1390                 }
1391         }
1392         write_unlock_bh(&tbl->lock);
1393         NEIGH_PRINTK1("neigh_parms_release: not found\n");
1394 }
1395 EXPORT_SYMBOL(neigh_parms_release);
1396
1397 static void neigh_parms_destroy(struct neigh_parms *parms)
1398 {
1399         release_net(neigh_parms_net(parms));
1400         kfree(parms);
1401 }
1402
1403 static struct lock_class_key neigh_table_proxy_queue_class;
1404
1405 void neigh_table_init_no_netlink(struct neigh_table *tbl)
1406 {
1407         unsigned long now = jiffies;
1408         unsigned long phsize;
1409
1410 #ifdef CONFIG_NET_NS
1411         tbl->parms.net = &init_net;
1412 #endif
1413         atomic_set(&tbl->parms.refcnt, 1);
1414         tbl->parms.reachable_time =
1415                           neigh_rand_reach_time(tbl->parms.base_reachable_time);
1416
1417         if (!tbl->kmem_cachep)
1418                 tbl->kmem_cachep =
1419                         kmem_cache_create(tbl->id, tbl->entry_size, 0,
1420                                           SLAB_HWCACHE_ALIGN|SLAB_PANIC,
1421                                           NULL);
1422         tbl->stats = alloc_percpu(struct neigh_statistics);
1423         if (!tbl->stats)
1424                 panic("cannot create neighbour cache statistics");
1425
1426 #ifdef CONFIG_PROC_FS
1427         tbl->pde = proc_create_data(tbl->id, 0, init_net.proc_net_stat,
1428                                     &neigh_stat_seq_fops, tbl);
1429         if (!tbl->pde)
1430                 panic("cannot create neighbour proc dir entry");
1431 #endif
1432
1433         tbl->hash_mask = 1;
1434         tbl->hash_buckets = neigh_hash_alloc(tbl->hash_mask + 1);
1435
1436         phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *);
1437         tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL);
1438
1439         if (!tbl->hash_buckets || !tbl->phash_buckets)
1440                 panic("cannot allocate neighbour cache hashes");
1441
1442         get_random_bytes(&tbl->hash_rnd, sizeof(tbl->hash_rnd));
1443
1444         rwlock_init(&tbl->lock);
1445         setup_timer(&tbl->gc_timer, neigh_periodic_timer, (unsigned long)tbl);
1446         tbl->gc_timer.expires  = now + 1;
1447         add_timer(&tbl->gc_timer);
1448
1449         setup_timer(&tbl->proxy_timer, neigh_proxy_process, (unsigned long)tbl);
1450         skb_queue_head_init_class(&tbl->proxy_queue,
1451                         &neigh_table_proxy_queue_class);
1452
1453         tbl->last_flush = now;
1454         tbl->last_rand  = now + tbl->parms.reachable_time * 20;
1455 }
1456 EXPORT_SYMBOL(neigh_table_init_no_netlink);
1457
1458 void neigh_table_init(struct neigh_table *tbl)
1459 {
1460         struct neigh_table *tmp;
1461
1462         neigh_table_init_no_netlink(tbl);
1463         write_lock(&neigh_tbl_lock);
1464         for (tmp = neigh_tables; tmp; tmp = tmp->next) {
1465                 if (tmp->family == tbl->family)
1466                         break;
1467         }
1468         tbl->next       = neigh_tables;
1469         neigh_tables    = tbl;
1470         write_unlock(&neigh_tbl_lock);
1471
1472         if (unlikely(tmp)) {
1473                 printk(KERN_ERR "NEIGH: Registering multiple tables for "
1474                        "family %d\n", tbl->family);
1475                 dump_stack();
1476         }
1477 }
1478 EXPORT_SYMBOL(neigh_table_init);
1479
1480 int neigh_table_clear(struct neigh_table *tbl)
1481 {
1482         struct neigh_table **tp;
1483
1484         /* It is not clean... Fix it to unload IPv6 module safely */
1485         del_timer_sync(&tbl->gc_timer);
1486         del_timer_sync(&tbl->proxy_timer);
1487         pneigh_queue_purge(&tbl->proxy_queue);
1488         neigh_ifdown(tbl, NULL);
1489         if (atomic_read(&tbl->entries))
1490                 printk(KERN_CRIT "neighbour leakage\n");
1491         write_lock(&neigh_tbl_lock);
1492         for (tp = &neigh_tables; *tp; tp = &(*tp)->next) {
1493                 if (*tp == tbl) {
1494                         *tp = tbl->next;
1495                         break;
1496                 }
1497         }
1498         write_unlock(&neigh_tbl_lock);
1499
1500         neigh_hash_free(tbl->hash_buckets, tbl->hash_mask + 1);
1501         tbl->hash_buckets = NULL;
1502
1503         kfree(tbl->phash_buckets);
1504         tbl->phash_buckets = NULL;
1505
1506         remove_proc_entry(tbl->id, init_net.proc_net_stat);
1507
1508         free_percpu(tbl->stats);
1509         tbl->stats = NULL;
1510
1511         kmem_cache_destroy(tbl->kmem_cachep);
1512         tbl->kmem_cachep = NULL;
1513
1514         return 0;
1515 }
1516 EXPORT_SYMBOL(neigh_table_clear);
1517
1518 static int neigh_delete(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
1519 {
1520         struct net *net = sock_net(skb->sk);
1521         struct ndmsg *ndm;
1522         struct nlattr *dst_attr;
1523         struct neigh_table *tbl;
1524         struct net_device *dev = NULL;
1525         int err = -EINVAL;
1526
1527         if (nlmsg_len(nlh) < sizeof(*ndm))
1528                 goto out;
1529
1530         dst_attr = nlmsg_find_attr(nlh, sizeof(*ndm), NDA_DST);
1531         if (dst_attr == NULL)
1532                 goto out;
1533
1534         ndm = nlmsg_data(nlh);
1535         if (ndm->ndm_ifindex) {
1536                 dev = dev_get_by_index(net, ndm->ndm_ifindex);
1537                 if (dev == NULL) {
1538                         err = -ENODEV;
1539                         goto out;
1540                 }
1541         }
1542
1543         read_lock(&neigh_tbl_lock);
1544         for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1545                 struct neighbour *neigh;
1546
1547                 if (tbl->family != ndm->ndm_family)
1548                         continue;
1549                 read_unlock(&neigh_tbl_lock);
1550
1551                 if (nla_len(dst_attr) < tbl->key_len)
1552                         goto out_dev_put;
1553
1554                 if (ndm->ndm_flags & NTF_PROXY) {
1555                         err = pneigh_delete(tbl, net, nla_data(dst_attr), dev);
1556                         goto out_dev_put;
1557                 }
1558
1559                 if (dev == NULL)
1560                         goto out_dev_put;
1561
1562                 neigh = neigh_lookup(tbl, nla_data(dst_attr), dev);
1563                 if (neigh == NULL) {
1564                         err = -ENOENT;
1565                         goto out_dev_put;
1566                 }
1567
1568                 err = neigh_update(neigh, NULL, NUD_FAILED,
1569                                    NEIGH_UPDATE_F_OVERRIDE |
1570                                    NEIGH_UPDATE_F_ADMIN);
1571                 neigh_release(neigh);
1572                 goto out_dev_put;
1573         }
1574         read_unlock(&neigh_tbl_lock);
1575         err = -EAFNOSUPPORT;
1576
1577 out_dev_put:
1578         if (dev)
1579                 dev_put(dev);
1580 out:
1581         return err;
1582 }
1583
1584 static int neigh_add(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
1585 {
1586         struct net *net = sock_net(skb->sk);
1587         struct ndmsg *ndm;
1588         struct nlattr *tb[NDA_MAX+1];
1589         struct neigh_table *tbl;
1590         struct net_device *dev = NULL;
1591         int err;
1592
1593         err = nlmsg_parse(nlh, sizeof(*ndm), tb, NDA_MAX, NULL);
1594         if (err < 0)
1595                 goto out;
1596
1597         err = -EINVAL;
1598         if (tb[NDA_DST] == NULL)
1599                 goto out;
1600
1601         ndm = nlmsg_data(nlh);
1602         if (ndm->ndm_ifindex) {
1603                 dev = dev_get_by_index(net, ndm->ndm_ifindex);
1604                 if (dev == NULL) {
1605                         err = -ENODEV;
1606                         goto out;
1607                 }
1608
1609                 if (tb[NDA_LLADDR] && nla_len(tb[NDA_LLADDR]) < dev->addr_len)
1610                         goto out_dev_put;
1611         }
1612
1613         read_lock(&neigh_tbl_lock);
1614         for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1615                 int flags = NEIGH_UPDATE_F_ADMIN | NEIGH_UPDATE_F_OVERRIDE;
1616                 struct neighbour *neigh;
1617                 void *dst, *lladdr;
1618
1619                 if (tbl->family != ndm->ndm_family)
1620                         continue;
1621                 read_unlock(&neigh_tbl_lock);
1622
1623                 if (nla_len(tb[NDA_DST]) < tbl->key_len)
1624                         goto out_dev_put;
1625                 dst = nla_data(tb[NDA_DST]);
1626                 lladdr = tb[NDA_LLADDR] ? nla_data(tb[NDA_LLADDR]) : NULL;
1627
1628                 if (ndm->ndm_flags & NTF_PROXY) {
1629                         struct pneigh_entry *pn;
1630
1631                         err = -ENOBUFS;
1632                         pn = pneigh_lookup(tbl, net, dst, dev, 1);
1633                         if (pn) {
1634                                 pn->flags = ndm->ndm_flags;
1635                                 err = 0;
1636                         }
1637                         goto out_dev_put;
1638                 }
1639
1640                 if (dev == NULL)
1641                         goto out_dev_put;
1642
1643                 neigh = neigh_lookup(tbl, dst, dev);
1644                 if (neigh == NULL) {
1645                         if (!(nlh->nlmsg_flags & NLM_F_CREATE)) {
1646                                 err = -ENOENT;
1647                                 goto out_dev_put;
1648                         }
1649
1650                         neigh = __neigh_lookup_errno(tbl, dst, dev);
1651                         if (IS_ERR(neigh)) {
1652                                 err = PTR_ERR(neigh);
1653                                 goto out_dev_put;
1654                         }
1655                 } else {
1656                         if (nlh->nlmsg_flags & NLM_F_EXCL) {
1657                                 err = -EEXIST;
1658                                 neigh_release(neigh);
1659                                 goto out_dev_put;
1660                         }
1661
1662                         if (!(nlh->nlmsg_flags & NLM_F_REPLACE))
1663                                 flags &= ~NEIGH_UPDATE_F_OVERRIDE;
1664                 }
1665
1666                 err = neigh_update(neigh, lladdr, ndm->ndm_state, flags);
1667                 neigh_release(neigh);
1668                 goto out_dev_put;
1669         }
1670
1671         read_unlock(&neigh_tbl_lock);
1672         err = -EAFNOSUPPORT;
1673
1674 out_dev_put:
1675         if (dev)
1676                 dev_put(dev);
1677 out:
1678         return err;
1679 }
1680
1681 static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms)
1682 {
1683         struct nlattr *nest;
1684
1685         nest = nla_nest_start(skb, NDTA_PARMS);
1686         if (nest == NULL)
1687                 return -ENOBUFS;
1688
1689         if (parms->dev)
1690                 NLA_PUT_U32(skb, NDTPA_IFINDEX, parms->dev->ifindex);
1691
1692         NLA_PUT_U32(skb, NDTPA_REFCNT, atomic_read(&parms->refcnt));
1693         NLA_PUT_U32(skb, NDTPA_QUEUE_LEN, parms->queue_len);
1694         NLA_PUT_U32(skb, NDTPA_PROXY_QLEN, parms->proxy_qlen);
1695         NLA_PUT_U32(skb, NDTPA_APP_PROBES, parms->app_probes);
1696         NLA_PUT_U32(skb, NDTPA_UCAST_PROBES, parms->ucast_probes);
1697         NLA_PUT_U32(skb, NDTPA_MCAST_PROBES, parms->mcast_probes);
1698         NLA_PUT_MSECS(skb, NDTPA_REACHABLE_TIME, parms->reachable_time);
1699         NLA_PUT_MSECS(skb, NDTPA_BASE_REACHABLE_TIME,
1700                       parms->base_reachable_time);
1701         NLA_PUT_MSECS(skb, NDTPA_GC_STALETIME, parms->gc_staletime);
1702         NLA_PUT_MSECS(skb, NDTPA_DELAY_PROBE_TIME, parms->delay_probe_time);
1703         NLA_PUT_MSECS(skb, NDTPA_RETRANS_TIME, parms->retrans_time);
1704         NLA_PUT_MSECS(skb, NDTPA_ANYCAST_DELAY, parms->anycast_delay);
1705         NLA_PUT_MSECS(skb, NDTPA_PROXY_DELAY, parms->proxy_delay);
1706         NLA_PUT_MSECS(skb, NDTPA_LOCKTIME, parms->locktime);
1707
1708         return nla_nest_end(skb, nest);
1709
1710 nla_put_failure:
1711         nla_nest_cancel(skb, nest);
1712         return -EMSGSIZE;
1713 }
1714
1715 static int neightbl_fill_info(struct sk_buff *skb, struct neigh_table *tbl,
1716                               u32 pid, u32 seq, int type, int flags)
1717 {
1718         struct nlmsghdr *nlh;
1719         struct ndtmsg *ndtmsg;
1720
1721         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1722         if (nlh == NULL)
1723                 return -EMSGSIZE;
1724
1725         ndtmsg = nlmsg_data(nlh);
1726
1727         read_lock_bh(&tbl->lock);
1728         ndtmsg->ndtm_family = tbl->family;
1729         ndtmsg->ndtm_pad1   = 0;
1730         ndtmsg->ndtm_pad2   = 0;
1731
1732         NLA_PUT_STRING(skb, NDTA_NAME, tbl->id);
1733         NLA_PUT_MSECS(skb, NDTA_GC_INTERVAL, tbl->gc_interval);
1734         NLA_PUT_U32(skb, NDTA_THRESH1, tbl->gc_thresh1);
1735         NLA_PUT_U32(skb, NDTA_THRESH2, tbl->gc_thresh2);
1736         NLA_PUT_U32(skb, NDTA_THRESH3, tbl->gc_thresh3);
1737
1738         {
1739                 unsigned long now = jiffies;
1740                 unsigned int flush_delta = now - tbl->last_flush;
1741                 unsigned int rand_delta = now - tbl->last_rand;
1742
1743                 struct ndt_config ndc = {
1744                         .ndtc_key_len           = tbl->key_len,
1745                         .ndtc_entry_size        = tbl->entry_size,
1746                         .ndtc_entries           = atomic_read(&tbl->entries),
1747                         .ndtc_last_flush        = jiffies_to_msecs(flush_delta),
1748                         .ndtc_last_rand         = jiffies_to_msecs(rand_delta),
1749                         .ndtc_hash_rnd          = tbl->hash_rnd,
1750                         .ndtc_hash_mask         = tbl->hash_mask,
1751                         .ndtc_hash_chain_gc     = tbl->hash_chain_gc,
1752                         .ndtc_proxy_qlen        = tbl->proxy_queue.qlen,
1753                 };
1754
1755                 NLA_PUT(skb, NDTA_CONFIG, sizeof(ndc), &ndc);
1756         }
1757
1758         {
1759                 int cpu;
1760                 struct ndt_stats ndst;
1761
1762                 memset(&ndst, 0, sizeof(ndst));
1763
1764                 for_each_possible_cpu(cpu) {
1765                         struct neigh_statistics *st;
1766
1767                         st = per_cpu_ptr(tbl->stats, cpu);
1768                         ndst.ndts_allocs                += st->allocs;
1769                         ndst.ndts_destroys              += st->destroys;
1770                         ndst.ndts_hash_grows            += st->hash_grows;
1771                         ndst.ndts_res_failed            += st->res_failed;
1772                         ndst.ndts_lookups               += st->lookups;
1773                         ndst.ndts_hits                  += st->hits;
1774                         ndst.ndts_rcv_probes_mcast      += st->rcv_probes_mcast;
1775                         ndst.ndts_rcv_probes_ucast      += st->rcv_probes_ucast;
1776                         ndst.ndts_periodic_gc_runs      += st->periodic_gc_runs;
1777                         ndst.ndts_forced_gc_runs        += st->forced_gc_runs;
1778                 }
1779
1780                 NLA_PUT(skb, NDTA_STATS, sizeof(ndst), &ndst);
1781         }
1782
1783         BUG_ON(tbl->parms.dev);
1784         if (neightbl_fill_parms(skb, &tbl->parms) < 0)
1785                 goto nla_put_failure;
1786
1787         read_unlock_bh(&tbl->lock);
1788         return nlmsg_end(skb, nlh);
1789
1790 nla_put_failure:
1791         read_unlock_bh(&tbl->lock);
1792         nlmsg_cancel(skb, nlh);
1793         return -EMSGSIZE;
1794 }
1795
1796 static int neightbl_fill_param_info(struct sk_buff *skb,
1797                                     struct neigh_table *tbl,
1798                                     struct neigh_parms *parms,
1799                                     u32 pid, u32 seq, int type,
1800                                     unsigned int flags)
1801 {
1802         struct ndtmsg *ndtmsg;
1803         struct nlmsghdr *nlh;
1804
1805         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1806         if (nlh == NULL)
1807                 return -EMSGSIZE;
1808
1809         ndtmsg = nlmsg_data(nlh);
1810
1811         read_lock_bh(&tbl->lock);
1812         ndtmsg->ndtm_family = tbl->family;
1813         ndtmsg->ndtm_pad1   = 0;
1814         ndtmsg->ndtm_pad2   = 0;
1815
1816         if (nla_put_string(skb, NDTA_NAME, tbl->id) < 0 ||
1817             neightbl_fill_parms(skb, parms) < 0)
1818                 goto errout;
1819
1820         read_unlock_bh(&tbl->lock);
1821         return nlmsg_end(skb, nlh);
1822 errout:
1823         read_unlock_bh(&tbl->lock);
1824         nlmsg_cancel(skb, nlh);
1825         return -EMSGSIZE;
1826 }
1827
1828 static const struct nla_policy nl_neightbl_policy[NDTA_MAX+1] = {
1829         [NDTA_NAME]             = { .type = NLA_STRING },
1830         [NDTA_THRESH1]          = { .type = NLA_U32 },
1831         [NDTA_THRESH2]          = { .type = NLA_U32 },
1832         [NDTA_THRESH3]          = { .type = NLA_U32 },
1833         [NDTA_GC_INTERVAL]      = { .type = NLA_U64 },
1834         [NDTA_PARMS]            = { .type = NLA_NESTED },
1835 };
1836
1837 static const struct nla_policy nl_ntbl_parm_policy[NDTPA_MAX+1] = {
1838         [NDTPA_IFINDEX]                 = { .type = NLA_U32 },
1839         [NDTPA_QUEUE_LEN]               = { .type = NLA_U32 },
1840         [NDTPA_PROXY_QLEN]              = { .type = NLA_U32 },
1841         [NDTPA_APP_PROBES]              = { .type = NLA_U32 },
1842         [NDTPA_UCAST_PROBES]            = { .type = NLA_U32 },
1843         [NDTPA_MCAST_PROBES]            = { .type = NLA_U32 },
1844         [NDTPA_BASE_REACHABLE_TIME]     = { .type = NLA_U64 },
1845         [NDTPA_GC_STALETIME]            = { .type = NLA_U64 },
1846         [NDTPA_DELAY_PROBE_TIME]        = { .type = NLA_U64 },
1847         [NDTPA_RETRANS_TIME]            = { .type = NLA_U64 },
1848         [NDTPA_ANYCAST_DELAY]           = { .type = NLA_U64 },
1849         [NDTPA_PROXY_DELAY]             = { .type = NLA_U64 },
1850         [NDTPA_LOCKTIME]                = { .type = NLA_U64 },
1851 };
1852
1853 static int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
1854 {
1855         struct net *net = sock_net(skb->sk);
1856         struct neigh_table *tbl;
1857         struct ndtmsg *ndtmsg;
1858         struct nlattr *tb[NDTA_MAX+1];
1859         int err;
1860
1861         err = nlmsg_parse(nlh, sizeof(*ndtmsg), tb, NDTA_MAX,
1862                           nl_neightbl_policy);
1863         if (err < 0)
1864                 goto errout;
1865
1866         if (tb[NDTA_NAME] == NULL) {
1867                 err = -EINVAL;
1868                 goto errout;
1869         }
1870
1871         ndtmsg = nlmsg_data(nlh);
1872         read_lock(&neigh_tbl_lock);
1873         for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1874                 if (ndtmsg->ndtm_family && tbl->family != ndtmsg->ndtm_family)
1875                         continue;
1876
1877                 if (nla_strcmp(tb[NDTA_NAME], tbl->id) == 0)
1878                         break;
1879         }
1880
1881         if (tbl == NULL) {
1882                 err = -ENOENT;
1883                 goto errout_locked;
1884         }
1885
1886         /*
1887          * We acquire tbl->lock to be nice to the periodic timers and
1888          * make sure they always see a consistent set of values.
1889          */
1890         write_lock_bh(&tbl->lock);
1891
1892         if (tb[NDTA_PARMS]) {
1893                 struct nlattr *tbp[NDTPA_MAX+1];
1894                 struct neigh_parms *p;
1895                 int i, ifindex = 0;
1896
1897                 err = nla_parse_nested(tbp, NDTPA_MAX, tb[NDTA_PARMS],
1898                                        nl_ntbl_parm_policy);
1899                 if (err < 0)
1900                         goto errout_tbl_lock;
1901
1902                 if (tbp[NDTPA_IFINDEX])
1903                         ifindex = nla_get_u32(tbp[NDTPA_IFINDEX]);
1904
1905                 p = lookup_neigh_params(tbl, net, ifindex);
1906                 if (p == NULL) {
1907                         err = -ENOENT;
1908                         goto errout_tbl_lock;
1909                 }
1910
1911                 for (i = 1; i <= NDTPA_MAX; i++) {
1912                         if (tbp[i] == NULL)
1913                                 continue;
1914
1915                         switch (i) {
1916                         case NDTPA_QUEUE_LEN:
1917                                 p->queue_len = nla_get_u32(tbp[i]);
1918                                 break;
1919                         case NDTPA_PROXY_QLEN:
1920                                 p->proxy_qlen = nla_get_u32(tbp[i]);
1921                                 break;
1922                         case NDTPA_APP_PROBES:
1923                                 p->app_probes = nla_get_u32(tbp[i]);
1924                                 break;
1925                         case NDTPA_UCAST_PROBES:
1926                                 p->ucast_probes = nla_get_u32(tbp[i]);
1927                                 break;
1928                         case NDTPA_MCAST_PROBES:
1929                                 p->mcast_probes = nla_get_u32(tbp[i]);
1930                                 break;
1931                         case NDTPA_BASE_REACHABLE_TIME:
1932                                 p->base_reachable_time = nla_get_msecs(tbp[i]);
1933                                 break;
1934                         case NDTPA_GC_STALETIME:
1935                                 p->gc_staletime = nla_get_msecs(tbp[i]);
1936                                 break;
1937                         case NDTPA_DELAY_PROBE_TIME:
1938                                 p->delay_probe_time = nla_get_msecs(tbp[i]);
1939                                 break;
1940                         case NDTPA_RETRANS_TIME:
1941                                 p->retrans_time = nla_get_msecs(tbp[i]);
1942                                 break;
1943                         case NDTPA_ANYCAST_DELAY:
1944                                 p->anycast_delay = nla_get_msecs(tbp[i]);
1945                                 break;
1946                         case NDTPA_PROXY_DELAY:
1947                                 p->proxy_delay = nla_get_msecs(tbp[i]);
1948                                 break;
1949                         case NDTPA_LOCKTIME:
1950                                 p->locktime = nla_get_msecs(tbp[i]);
1951                                 break;
1952                         }
1953                 }
1954         }
1955
1956         if (tb[NDTA_THRESH1])
1957                 tbl->gc_thresh1 = nla_get_u32(tb[NDTA_THRESH1]);
1958
1959         if (tb[NDTA_THRESH2])
1960                 tbl->gc_thresh2 = nla_get_u32(tb[NDTA_THRESH2]);
1961
1962         if (tb[NDTA_THRESH3])
1963                 tbl->gc_thresh3 = nla_get_u32(tb[NDTA_THRESH3]);
1964
1965         if (tb[NDTA_GC_INTERVAL])
1966                 tbl->gc_interval = nla_get_msecs(tb[NDTA_GC_INTERVAL]);
1967
1968         err = 0;
1969
1970 errout_tbl_lock:
1971         write_unlock_bh(&tbl->lock);
1972 errout_locked:
1973         read_unlock(&neigh_tbl_lock);
1974 errout:
1975         return err;
1976 }
1977
1978 static int neightbl_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
1979 {
1980         struct net *net = sock_net(skb->sk);
1981         int family, tidx, nidx = 0;
1982         int tbl_skip = cb->args[0];
1983         int neigh_skip = cb->args[1];
1984         struct neigh_table *tbl;
1985
1986         family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
1987
1988         read_lock(&neigh_tbl_lock);
1989         for (tbl = neigh_tables, tidx = 0; tbl; tbl = tbl->next, tidx++) {
1990                 struct neigh_parms *p;
1991
1992                 if (tidx < tbl_skip || (family && tbl->family != family))
1993                         continue;
1994
1995                 if (neightbl_fill_info(skb, tbl, NETLINK_CB(cb->skb).pid,
1996                                        cb->nlh->nlmsg_seq, RTM_NEWNEIGHTBL,
1997                                        NLM_F_MULTI) <= 0)
1998                         break;
1999
2000                 for (nidx = 0, p = tbl->parms.next; p; p = p->next) {
2001                         if (!net_eq(neigh_parms_net(p), net))
2002                                 continue;
2003
2004                         if (nidx++ < neigh_skip)
2005                                 continue;
2006
2007                         if (neightbl_fill_param_info(skb, tbl, p,
2008                                                      NETLINK_CB(cb->skb).pid,
2009                                                      cb->nlh->nlmsg_seq,
2010                                                      RTM_NEWNEIGHTBL,
2011                                                      NLM_F_MULTI) <= 0)
2012                                 goto out;
2013                 }
2014
2015                 neigh_skip = 0;
2016         }
2017 out:
2018         read_unlock(&neigh_tbl_lock);
2019         cb->args[0] = tidx;
2020         cb->args[1] = nidx;
2021
2022         return skb->len;
2023 }
2024
2025 static int neigh_fill_info(struct sk_buff *skb, struct neighbour *neigh,
2026                            u32 pid, u32 seq, int type, unsigned int flags)
2027 {
2028         unsigned long now = jiffies;
2029         struct nda_cacheinfo ci;
2030         struct nlmsghdr *nlh;
2031         struct ndmsg *ndm;
2032
2033         nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2034         if (nlh == NULL)
2035                 return -EMSGSIZE;
2036
2037         ndm = nlmsg_data(nlh);
2038         ndm->ndm_family  = neigh->ops->family;
2039         ndm->ndm_pad1    = 0;
2040         ndm->ndm_pad2    = 0;
2041         ndm->ndm_flags   = neigh->flags;
2042         ndm->ndm_type    = neigh->type;
2043         ndm->ndm_ifindex = neigh->dev->ifindex;
2044
2045         NLA_PUT(skb, NDA_DST, neigh->tbl->key_len, neigh->primary_key);
2046
2047         read_lock_bh(&neigh->lock);
2048         ndm->ndm_state   = neigh->nud_state;
2049         if ((neigh->nud_state & NUD_VALID) &&
2050             nla_put(skb, NDA_LLADDR, neigh->dev->addr_len, neigh->ha) < 0) {
2051                 read_unlock_bh(&neigh->lock);
2052                 goto nla_put_failure;
2053         }
2054
2055         ci.ndm_used      = jiffies_to_clock_t(now - neigh->used);
2056         ci.ndm_confirmed = jiffies_to_clock_t(now - neigh->confirmed);
2057         ci.ndm_updated   = jiffies_to_clock_t(now - neigh->updated);
2058         ci.ndm_refcnt    = atomic_read(&neigh->refcnt) - 1;
2059         read_unlock_bh(&neigh->lock);
2060
2061         NLA_PUT_U32(skb, NDA_PROBES, atomic_read(&neigh->probes));
2062         NLA_PUT(skb, NDA_CACHEINFO, sizeof(ci), &ci);
2063
2064         return nlmsg_end(skb, nlh);
2065
2066 nla_put_failure:
2067         nlmsg_cancel(skb, nlh);
2068         return -EMSGSIZE;
2069 }
2070
2071 static void neigh_update_notify(struct neighbour *neigh)
2072 {
2073         call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
2074         __neigh_notify(neigh, RTM_NEWNEIGH, 0);
2075 }
2076
2077 static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2078                             struct netlink_callback *cb)
2079 {
2080         struct net * net = sock_net(skb->sk);
2081         struct neighbour *n;
2082         int rc, h, s_h = cb->args[1];
2083         int idx, s_idx = idx = cb->args[2];
2084
2085         read_lock_bh(&tbl->lock);
2086         for (h = 0; h <= tbl->hash_mask; h++) {
2087                 if (h < s_h)
2088                         continue;
2089                 if (h > s_h)
2090                         s_idx = 0;
2091                 for (n = tbl->hash_buckets[h], idx = 0; n; n = n->next) {
2092                         int lidx;
2093                         if (dev_net(n->dev) != net)
2094                                 continue;
2095                         lidx = idx++;
2096                         if (lidx < s_idx)
2097                                 continue;
2098                         if (neigh_fill_info(skb, n, NETLINK_CB(cb->skb).pid,
2099                                             cb->nlh->nlmsg_seq,
2100                                             RTM_NEWNEIGH,
2101                                             NLM_F_MULTI) <= 0) {
2102                                 read_unlock_bh(&tbl->lock);
2103                                 rc = -1;
2104                                 goto out;
2105                         }
2106                 }
2107         }
2108         read_unlock_bh(&tbl->lock);
2109         rc = skb->len;
2110 out:
2111         cb->args[1] = h;
2112         cb->args[2] = idx;
2113         return rc;
2114 }
2115
2116 static int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2117 {
2118         struct neigh_table *tbl;
2119         int t, family, s_t;
2120
2121         read_lock(&neigh_tbl_lock);
2122         family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2123         s_t = cb->args[0];
2124
2125         for (tbl = neigh_tables, t = 0; tbl; tbl = tbl->next, t++) {
2126                 if (t < s_t || (family && tbl->family != family))
2127                         continue;
2128                 if (t > s_t)
2129                         memset(&cb->args[1], 0, sizeof(cb->args) -
2130                                                 sizeof(cb->args[0]));
2131                 if (neigh_dump_table(tbl, skb, cb) < 0)
2132                         break;
2133         }
2134         read_unlock(&neigh_tbl_lock);
2135
2136         cb->args[0] = t;
2137         return skb->len;
2138 }
2139
2140 void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie)
2141 {
2142         int chain;
2143
2144         read_lock_bh(&tbl->lock);
2145         for (chain = 0; chain <= tbl->hash_mask; chain++) {
2146                 struct neighbour *n;
2147
2148                 for (n = tbl->hash_buckets[chain]; n; n = n->next)
2149                         cb(n, cookie);
2150         }
2151         read_unlock_bh(&tbl->lock);
2152 }
2153 EXPORT_SYMBOL(neigh_for_each);
2154
2155 /* The tbl->lock must be held as a writer and BH disabled. */
2156 void __neigh_for_each_release(struct neigh_table *tbl,
2157                               int (*cb)(struct neighbour *))
2158 {
2159         int chain;
2160
2161         for (chain = 0; chain <= tbl->hash_mask; chain++) {
2162                 struct neighbour *n, **np;
2163
2164                 np = &tbl->hash_buckets[chain];
2165                 while ((n = *np) != NULL) {
2166                         int release;
2167
2168                         write_lock(&n->lock);
2169                         release = cb(n);
2170                         if (release) {
2171                                 *np = n->next;
2172                                 n->dead = 1;
2173                         } else
2174                                 np = &n->next;
2175                         write_unlock(&n->lock);
2176                         if (release)
2177                                 neigh_cleanup_and_release(n);
2178                 }
2179         }
2180 }
2181 EXPORT_SYMBOL(__neigh_for_each_release);
2182
2183 #ifdef CONFIG_PROC_FS
2184
2185 static struct neighbour *neigh_get_first(struct seq_file *seq)
2186 {
2187         struct neigh_seq_state *state = seq->private;
2188         struct net *net = seq_file_net(seq);
2189         struct neigh_table *tbl = state->tbl;
2190         struct neighbour *n = NULL;
2191         int bucket = state->bucket;
2192
2193         state->flags &= ~NEIGH_SEQ_IS_PNEIGH;
2194         for (bucket = 0; bucket <= tbl->hash_mask; bucket++) {
2195                 n = tbl->hash_buckets[bucket];
2196
2197                 while (n) {
2198                         if (!net_eq(dev_net(n->dev), net))
2199                                 goto next;
2200                         if (state->neigh_sub_iter) {
2201                                 loff_t fakep = 0;
2202                                 void *v;
2203
2204                                 v = state->neigh_sub_iter(state, n, &fakep);
2205                                 if (!v)
2206                                         goto next;
2207                         }
2208                         if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2209                                 break;
2210                         if (n->nud_state & ~NUD_NOARP)
2211                                 break;
2212                 next:
2213                         n = n->next;
2214                 }
2215
2216                 if (n)
2217                         break;
2218         }
2219         state->bucket = bucket;
2220
2221         return n;
2222 }
2223
2224 static struct neighbour *neigh_get_next(struct seq_file *seq,
2225                                         struct neighbour *n,
2226                                         loff_t *pos)
2227 {
2228         struct neigh_seq_state *state = seq->private;
2229         struct net *net = seq_file_net(seq);
2230         struct neigh_table *tbl = state->tbl;
2231
2232         if (state->neigh_sub_iter) {
2233                 void *v = state->neigh_sub_iter(state, n, pos);
2234                 if (v)
2235                         return n;
2236         }
2237         n = n->next;
2238
2239         while (1) {
2240                 while (n) {
2241                         if (!net_eq(dev_net(n->dev), net))
2242                                 goto next;
2243                         if (state->neigh_sub_iter) {
2244                                 void *v = state->neigh_sub_iter(state, n, pos);
2245                                 if (v)
2246                                         return n;
2247                                 goto next;
2248                         }
2249                         if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2250                                 break;
2251
2252                         if (n->nud_state & ~NUD_NOARP)
2253                                 break;
2254                 next:
2255                         n = n->next;
2256                 }
2257
2258                 if (n)
2259                         break;
2260
2261                 if (++state->bucket > tbl->hash_mask)
2262                         break;
2263
2264                 n = tbl->hash_buckets[state->bucket];
2265         }
2266
2267         if (n && pos)
2268                 --(*pos);
2269         return n;
2270 }
2271
2272 static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos)
2273 {
2274         struct neighbour *n = neigh_get_first(seq);
2275
2276         if (n) {
2277                 --(*pos);
2278                 while (*pos) {
2279                         n = neigh_get_next(seq, n, pos);
2280                         if (!n)
2281                                 break;
2282                 }
2283         }
2284         return *pos ? NULL : n;
2285 }
2286
2287 static struct pneigh_entry *pneigh_get_first(struct seq_file *seq)
2288 {
2289         struct neigh_seq_state *state = seq->private;
2290         struct net *net = seq_file_net(seq);
2291         struct neigh_table *tbl = state->tbl;
2292         struct pneigh_entry *pn = NULL;
2293         int bucket = state->bucket;
2294
2295         state->flags |= NEIGH_SEQ_IS_PNEIGH;
2296         for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) {
2297                 pn = tbl->phash_buckets[bucket];
2298                 while (pn && !net_eq(pneigh_net(pn), net))
2299                         pn = pn->next;
2300                 if (pn)
2301                         break;
2302         }
2303         state->bucket = bucket;
2304
2305         return pn;
2306 }
2307
2308 static struct pneigh_entry *pneigh_get_next(struct seq_file *seq,
2309                                             struct pneigh_entry *pn,
2310                                             loff_t *pos)
2311 {
2312         struct neigh_seq_state *state = seq->private;
2313         struct net *net = seq_file_net(seq);
2314         struct neigh_table *tbl = state->tbl;
2315
2316         pn = pn->next;
2317         while (!pn) {
2318                 if (++state->bucket > PNEIGH_HASHMASK)
2319                         break;
2320                 pn = tbl->phash_buckets[state->bucket];
2321                 while (pn && !net_eq(pneigh_net(pn), net))
2322                         pn = pn->next;
2323                 if (pn)
2324                         break;
2325         }
2326
2327         if (pn && pos)
2328                 --(*pos);
2329
2330         return pn;
2331 }
2332
2333 static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos)
2334 {
2335         struct pneigh_entry *pn = pneigh_get_first(seq);
2336
2337         if (pn) {
2338                 --(*pos);
2339                 while (*pos) {
2340                         pn = pneigh_get_next(seq, pn, pos);
2341                         if (!pn)
2342                                 break;
2343                 }
2344         }
2345         return *pos ? NULL : pn;
2346 }
2347
2348 static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos)
2349 {
2350         struct neigh_seq_state *state = seq->private;
2351         void *rc;
2352         loff_t idxpos = *pos;
2353
2354         rc = neigh_get_idx(seq, &idxpos);
2355         if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2356                 rc = pneigh_get_idx(seq, &idxpos);
2357
2358         return rc;
2359 }
2360
2361 void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags)
2362         __acquires(tbl->lock)
2363 {
2364         struct neigh_seq_state *state = seq->private;
2365
2366         state->tbl = tbl;
2367         state->bucket = 0;
2368         state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH);
2369
2370         read_lock_bh(&tbl->lock);
2371
2372         return *pos ? neigh_get_idx_any(seq, pos) : SEQ_START_TOKEN;
2373 }
2374 EXPORT_SYMBOL(neigh_seq_start);
2375
2376 void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2377 {
2378         struct neigh_seq_state *state;
2379         void *rc;
2380
2381         if (v == SEQ_START_TOKEN) {
2382                 rc = neigh_get_first(seq);
2383                 goto out;
2384         }
2385
2386         state = seq->private;
2387         if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) {
2388                 rc = neigh_get_next(seq, v, NULL);
2389                 if (rc)
2390                         goto out;
2391                 if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2392                         rc = pneigh_get_first(seq);
2393         } else {
2394                 BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY);
2395                 rc = pneigh_get_next(seq, v, NULL);
2396         }
2397 out:
2398         ++(*pos);
2399         return rc;
2400 }
2401 EXPORT_SYMBOL(neigh_seq_next);
2402
2403 void neigh_seq_stop(struct seq_file *seq, void *v)
2404         __releases(tbl->lock)
2405 {
2406         struct neigh_seq_state *state = seq->private;
2407         struct neigh_table *tbl = state->tbl;
2408
2409         read_unlock_bh(&tbl->lock);
2410 }
2411 EXPORT_SYMBOL(neigh_seq_stop);
2412
2413 /* statistics via seq_file */
2414
2415 static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos)
2416 {
2417         struct proc_dir_entry *pde = seq->private;
2418         struct neigh_table *tbl = pde->data;
2419         int cpu;
2420
2421         if (*pos == 0)
2422                 return SEQ_START_TOKEN;
2423
2424         for (cpu = *pos-1; cpu < NR_CPUS; ++cpu) {
2425                 if (!cpu_possible(cpu))
2426                         continue;
2427                 *pos = cpu+1;
2428                 return per_cpu_ptr(tbl->stats, cpu);
2429         }
2430         return NULL;
2431 }
2432
2433 static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2434 {
2435         struct proc_dir_entry *pde = seq->private;
2436         struct neigh_table *tbl = pde->data;
2437         int cpu;
2438
2439         for (cpu = *pos; cpu < NR_CPUS; ++cpu) {
2440                 if (!cpu_possible(cpu))
2441                         continue;
2442                 *pos = cpu+1;
2443                 return per_cpu_ptr(tbl->stats, cpu);
2444         }
2445         return NULL;
2446 }
2447
2448 static void neigh_stat_seq_stop(struct seq_file *seq, void *v)
2449 {
2450
2451 }
2452
2453 static int neigh_stat_seq_show(struct seq_file *seq, void *v)
2454 {
2455         struct proc_dir_entry *pde = seq->private;
2456         struct neigh_table *tbl = pde->data;
2457         struct neigh_statistics *st = v;
2458
2459         if (v == SEQ_START_TOKEN) {
2460                 seq_printf(seq, "entries  allocs destroys hash_grows  lookups hits  res_failed  rcv_probes_mcast rcv_probes_ucast  periodic_gc_runs forced_gc_runs unresolved_discards\n");
2461                 return 0;
2462         }
2463
2464         seq_printf(seq, "%08x  %08lx %08lx %08lx  %08lx %08lx  %08lx  "
2465                         "%08lx %08lx  %08lx %08lx %08lx\n",
2466                    atomic_read(&tbl->entries),
2467
2468                    st->allocs,
2469                    st->destroys,
2470                    st->hash_grows,
2471
2472                    st->lookups,
2473                    st->hits,
2474
2475                    st->res_failed,
2476
2477                    st->rcv_probes_mcast,
2478                    st->rcv_probes_ucast,
2479
2480                    st->periodic_gc_runs,
2481                    st->forced_gc_runs,
2482                    st->unres_discards
2483                    );
2484
2485         return 0;
2486 }
2487
2488 static const struct seq_operations neigh_stat_seq_ops = {
2489         .start  = neigh_stat_seq_start,
2490         .next   = neigh_stat_seq_next,
2491         .stop   = neigh_stat_seq_stop,
2492         .show   = neigh_stat_seq_show,
2493 };
2494
2495 static int neigh_stat_seq_open(struct inode *inode, struct file *file)
2496 {
2497         int ret = seq_open(file, &neigh_stat_seq_ops);
2498
2499         if (!ret) {
2500                 struct seq_file *sf = file->private_data;
2501                 sf->private = PDE(inode);
2502         }
2503         return ret;
2504 };
2505
2506 static const struct file_operations neigh_stat_seq_fops = {
2507         .owner   = THIS_MODULE,
2508         .open    = neigh_stat_seq_open,
2509         .read    = seq_read,
2510         .llseek  = seq_lseek,
2511         .release = seq_release,
2512 };
2513
2514 #endif /* CONFIG_PROC_FS */
2515
2516 static inline size_t neigh_nlmsg_size(void)
2517 {
2518         return NLMSG_ALIGN(sizeof(struct ndmsg))
2519                + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2520                + nla_total_size(MAX_ADDR_LEN) /* NDA_LLADDR */
2521                + nla_total_size(sizeof(struct nda_cacheinfo))
2522                + nla_total_size(4); /* NDA_PROBES */
2523 }
2524
2525 static void __neigh_notify(struct neighbour *n, int type, int flags)
2526 {
2527         struct net *net = dev_net(n->dev);
2528         struct sk_buff *skb;
2529         int err = -ENOBUFS;
2530
2531         skb = nlmsg_new(neigh_nlmsg_size(), GFP_ATOMIC);
2532         if (skb == NULL)
2533                 goto errout;
2534
2535         err = neigh_fill_info(skb, n, 0, 0, type, flags);
2536         if (err < 0) {
2537                 /* -EMSGSIZE implies BUG in neigh_nlmsg_size() */
2538                 WARN_ON(err == -EMSGSIZE);
2539                 kfree_skb(skb);
2540                 goto errout;
2541         }
2542         err = rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
2543 errout:
2544         if (err < 0)
2545                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
2546 }
2547
2548 #ifdef CONFIG_ARPD
2549 void neigh_app_ns(struct neighbour *n)
2550 {
2551         __neigh_notify(n, RTM_GETNEIGH, NLM_F_REQUEST);
2552 }
2553 EXPORT_SYMBOL(neigh_app_ns);
2554 #endif /* CONFIG_ARPD */
2555
2556 #ifdef CONFIG_SYSCTL
2557
2558 static struct neigh_sysctl_table {
2559         struct ctl_table_header *sysctl_header;
2560         struct ctl_table neigh_vars[__NET_NEIGH_MAX];
2561         char *dev_name;
2562 } neigh_sysctl_template __read_mostly = {
2563         .neigh_vars = {
2564                 {
2565                         .ctl_name       = NET_NEIGH_MCAST_SOLICIT,
2566                         .procname       = "mcast_solicit",
2567                         .maxlen         = sizeof(int),
2568                         .mode           = 0644,
2569                         .proc_handler   = &proc_dointvec,
2570                 },
2571                 {
2572                         .ctl_name       = NET_NEIGH_UCAST_SOLICIT,
2573                         .procname       = "ucast_solicit",
2574                         .maxlen         = sizeof(int),
2575                         .mode           = 0644,
2576                         .proc_handler   = &proc_dointvec,
2577                 },
2578                 {
2579                         .ctl_name       = NET_NEIGH_APP_SOLICIT,
2580                         .procname       = "app_solicit",
2581                         .maxlen         = sizeof(int),
2582                         .mode           = 0644,
2583                         .proc_handler   = &proc_dointvec,
2584                 },
2585                 {
2586                         .procname       = "retrans_time",
2587                         .maxlen         = sizeof(int),
2588                         .mode           = 0644,
2589                         .proc_handler   = &proc_dointvec_userhz_jiffies,
2590                 },
2591                 {
2592                         .ctl_name       = NET_NEIGH_REACHABLE_TIME,
2593                         .procname       = "base_reachable_time",
2594                         .maxlen         = sizeof(int),
2595                         .mode           = 0644,
2596                         .proc_handler   = &proc_dointvec_jiffies,
2597                         .strategy       = &sysctl_jiffies,
2598                 },
2599                 {
2600                         .ctl_name       = NET_NEIGH_DELAY_PROBE_TIME,
2601                         .procname       = "delay_first_probe_time",
2602                         .maxlen         = sizeof(int),
2603                         .mode           = 0644,
2604                         .proc_handler   = &proc_dointvec_jiffies,
2605                         .strategy       = &sysctl_jiffies,
2606                 },
2607                 {
2608                         .ctl_name       = NET_NEIGH_GC_STALE_TIME,
2609                         .procname       = "gc_stale_time",
2610                         .maxlen         = sizeof(int),
2611                         .mode           = 0644,
2612                         .proc_handler   = &proc_dointvec_jiffies,
2613                         .strategy       = &sysctl_jiffies,
2614                 },
2615                 {
2616                         .ctl_name       = NET_NEIGH_UNRES_QLEN,
2617                         .procname       = "unres_qlen",
2618                         .maxlen         = sizeof(int),
2619                         .mode           = 0644,
2620                         .proc_handler   = &proc_dointvec,
2621                 },
2622                 {
2623                         .ctl_name       = NET_NEIGH_PROXY_QLEN,
2624                         .procname       = "proxy_qlen",
2625                         .maxlen         = sizeof(int),
2626                         .mode           = 0644,
2627                         .proc_handler   = &proc_dointvec,
2628                 },
2629                 {
2630                         .procname       = "anycast_delay",
2631                         .maxlen         = sizeof(int),
2632                         .mode           = 0644,
2633                         .proc_handler   = &proc_dointvec_userhz_jiffies,
2634                 },
2635                 {
2636                         .procname       = "proxy_delay",
2637                         .maxlen         = sizeof(int),
2638                         .mode           = 0644,
2639                         .proc_handler   = &proc_dointvec_userhz_jiffies,
2640                 },
2641                 {
2642                         .procname       = "locktime",
2643                         .maxlen         = sizeof(int),
2644                         .mode           = 0644,
2645                         .proc_handler   = &proc_dointvec_userhz_jiffies,
2646                 },
2647                 {
2648                         .ctl_name       = NET_NEIGH_RETRANS_TIME_MS,
2649                         .procname       = "retrans_time_ms",
2650                         .maxlen         = sizeof(int),
2651                         .mode           = 0644,
2652                         .proc_handler   = &proc_dointvec_ms_jiffies,
2653                         .strategy       = &sysctl_ms_jiffies,
2654                 },
2655                 {
2656                         .ctl_name       = NET_NEIGH_REACHABLE_TIME_MS,
2657                         .procname       = "base_reachable_time_ms",
2658                         .maxlen         = sizeof(int),
2659                         .mode           = 0644,
2660                         .proc_handler   = &proc_dointvec_ms_jiffies,
2661                         .strategy       = &sysctl_ms_jiffies,
2662                 },
2663                 {
2664                         .ctl_name       = NET_NEIGH_GC_INTERVAL,
2665                         .procname       = "gc_interval",
2666                         .maxlen         = sizeof(int),
2667                         .mode           = 0644,
2668                         .proc_handler   = &proc_dointvec_jiffies,
2669                         .strategy       = &sysctl_jiffies,
2670                 },
2671                 {
2672                         .ctl_name       = NET_NEIGH_GC_THRESH1,
2673                         .procname       = "gc_thresh1",
2674                         .maxlen         = sizeof(int),
2675                         .mode           = 0644,
2676                         .proc_handler   = &proc_dointvec,
2677                 },
2678                 {
2679                         .ctl_name       = NET_NEIGH_GC_THRESH2,
2680                         .procname       = "gc_thresh2",
2681                         .maxlen         = sizeof(int),
2682                         .mode           = 0644,
2683                         .proc_handler   = &proc_dointvec,
2684                 },
2685                 {
2686                         .ctl_name       = NET_NEIGH_GC_THRESH3,
2687                         .procname       = "gc_thresh3",
2688                         .maxlen         = sizeof(int),
2689                         .mode           = 0644,
2690                         .proc_handler   = &proc_dointvec,
2691                 },
2692                 {},
2693         },
2694 };
2695
2696 int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
2697                           int p_id, int pdev_id, char *p_name,
2698                           proc_handler *handler, ctl_handler *strategy)
2699 {
2700         struct neigh_sysctl_table *t;
2701         const char *dev_name_source = NULL;
2702
2703 #define NEIGH_CTL_PATH_ROOT     0
2704 #define NEIGH_CTL_PATH_PROTO    1
2705 #define NEIGH_CTL_PATH_NEIGH    2
2706 #define NEIGH_CTL_PATH_DEV      3
2707
2708         struct ctl_path neigh_path[] = {
2709                 { .procname = "net",     .ctl_name = CTL_NET, },
2710                 { .procname = "proto",   .ctl_name = 0, },
2711                 { .procname = "neigh",   .ctl_name = 0, },
2712                 { .procname = "default", .ctl_name = NET_PROTO_CONF_DEFAULT, },
2713                 { },
2714         };
2715
2716         t = kmemdup(&neigh_sysctl_template, sizeof(*t), GFP_KERNEL);
2717         if (!t)
2718                 goto err;
2719
2720         t->neigh_vars[0].data  = &p->mcast_probes;
2721         t->neigh_vars[1].data  = &p->ucast_probes;
2722         t->neigh_vars[2].data  = &p->app_probes;
2723         t->neigh_vars[3].data  = &p->retrans_time;
2724         t->neigh_vars[4].data  = &p->base_reachable_time;
2725         t->neigh_vars[5].data  = &p->delay_probe_time;
2726         t->neigh_vars[6].data  = &p->gc_staletime;
2727         t->neigh_vars[7].data  = &p->queue_len;
2728         t->neigh_vars[8].data  = &p->proxy_qlen;
2729         t->neigh_vars[9].data  = &p->anycast_delay;
2730         t->neigh_vars[10].data = &p->proxy_delay;
2731         t->neigh_vars[11].data = &p->locktime;
2732         t->neigh_vars[12].data  = &p->retrans_time;
2733         t->neigh_vars[13].data  = &p->base_reachable_time;
2734
2735         if (dev) {
2736                 dev_name_source = dev->name;
2737                 neigh_path[NEIGH_CTL_PATH_DEV].ctl_name = dev->ifindex;
2738                 /* Terminate the table early */
2739                 memset(&t->neigh_vars[14], 0, sizeof(t->neigh_vars[14]));
2740         } else {
2741                 dev_name_source = neigh_path[NEIGH_CTL_PATH_DEV].procname;
2742                 t->neigh_vars[14].data = (int *)(p + 1);
2743                 t->neigh_vars[15].data = (int *)(p + 1) + 1;
2744                 t->neigh_vars[16].data = (int *)(p + 1) + 2;
2745                 t->neigh_vars[17].data = (int *)(p + 1) + 3;
2746         }
2747
2748
2749         if (handler || strategy) {
2750                 /* RetransTime */
2751                 t->neigh_vars[3].proc_handler = handler;
2752                 t->neigh_vars[3].strategy = strategy;
2753                 t->neigh_vars[3].extra1 = dev;
2754                 if (!strategy)
2755                         t->neigh_vars[3].ctl_name = CTL_UNNUMBERED;
2756                 /* ReachableTime */
2757                 t->neigh_vars[4].proc_handler = handler;
2758                 t->neigh_vars[4].strategy = strategy;
2759                 t->neigh_vars[4].extra1 = dev;
2760                 if (!strategy)
2761                         t->neigh_vars[4].ctl_name = CTL_UNNUMBERED;
2762                 /* RetransTime (in milliseconds)*/
2763                 t->neigh_vars[12].proc_handler = handler;
2764                 t->neigh_vars[12].strategy = strategy;
2765                 t->neigh_vars[12].extra1 = dev;
2766                 if (!strategy)
2767                         t->neigh_vars[12].ctl_name = CTL_UNNUMBERED;
2768                 /* ReachableTime (in milliseconds) */
2769                 t->neigh_vars[13].proc_handler = handler;
2770                 t->neigh_vars[13].strategy = strategy;
2771                 t->neigh_vars[13].extra1 = dev;
2772                 if (!strategy)
2773                         t->neigh_vars[13].ctl_name = CTL_UNNUMBERED;
2774         }
2775
2776         t->dev_name = kstrdup(dev_name_source, GFP_KERNEL);
2777         if (!t->dev_name)
2778                 goto free;
2779
2780         neigh_path[NEIGH_CTL_PATH_DEV].procname = t->dev_name;
2781         neigh_path[NEIGH_CTL_PATH_NEIGH].ctl_name = pdev_id;
2782         neigh_path[NEIGH_CTL_PATH_PROTO].procname = p_name;
2783         neigh_path[NEIGH_CTL_PATH_PROTO].ctl_name = p_id;
2784
2785         t->sysctl_header =
2786                 register_net_sysctl_table(neigh_parms_net(p), neigh_path, t->neigh_vars);
2787         if (!t->sysctl_header)
2788                 goto free_procname;
2789
2790         p->sysctl_table = t;
2791         return 0;
2792
2793 free_procname:
2794         kfree(t->dev_name);
2795 free:
2796         kfree(t);
2797 err:
2798         return -ENOBUFS;
2799 }
2800 EXPORT_SYMBOL(neigh_sysctl_register);
2801
2802 void neigh_sysctl_unregister(struct neigh_parms *p)
2803 {
2804         if (p->sysctl_table) {
2805                 struct neigh_sysctl_table *t = p->sysctl_table;
2806                 p->sysctl_table = NULL;
2807                 unregister_sysctl_table(t->sysctl_header);
2808                 kfree(t->dev_name);
2809                 kfree(t);
2810         }
2811 }
2812 EXPORT_SYMBOL(neigh_sysctl_unregister);
2813
2814 #endif  /* CONFIG_SYSCTL */
2815
2816 static int __init neigh_init(void)
2817 {
2818         rtnl_register(PF_UNSPEC, RTM_NEWNEIGH, neigh_add, NULL);
2819         rtnl_register(PF_UNSPEC, RTM_DELNEIGH, neigh_delete, NULL);
2820         rtnl_register(PF_UNSPEC, RTM_GETNEIGH, NULL, neigh_dump_info);
2821
2822         rtnl_register(PF_UNSPEC, RTM_GETNEIGHTBL, NULL, neightbl_dump_info);
2823         rtnl_register(PF_UNSPEC, RTM_SETNEIGHTBL, neightbl_set, NULL);
2824
2825         return 0;
2826 }
2827
2828 subsys_initcall(neigh_init);
2829